@comchatx/cx-sponsored-ads-playbook 0.0.0-stage → 0.1.0

package/research/beyond/automation-and-amazon-tools.md

@@ -0,0 +1,725 @@
+# Automation and Amazon's own tools
+
+Read 2026-11-09 by the research seat. Tags: [official] Amazon's own words (Amazon Ads help, library, learn and news pages, Ads API guides, OpenAPI models, MCP tool specifications, release notes); [opinion] a named practitioner, agency or paid tool, said as theirs; [research] published research; [seat] the seat's own reasoning, allowed here, never a position. Every page below was opened on 2026-11-09 unless the line says otherwise. Figures (limits, windows, percentages, dates) are recorded with their page; a skill reads them at run time, and any figure marked *(for human readers only)* is not to be copied into a skill sentence.
+
+## What this covers
+
+This file maps what Amazon itself automates for an advertiser as of November 2026, what Amazon exposes to an AI agent, and what the paid tools sellers use automate and how they say they do it. It covers:
+
+- Amazon's automated bidding and its rules: the bidding strategies (named here, owned by `areas/bids-and-placements.md`), rule-based bidding, schedule bid rules, and the newer optimization-rules model that carries performance conditions;
+- Amazon's budget automation: schedule and performance budget rules, recommended events, portfolio unspent-budget sharing, budget recommendations (the mechanics of pacing belong to `areas/budgets-and-pacing.md`);
+- Amazon's recommendation surfaces and their "apply" flows: the consolidated Sponsored Products recommendations, theme-based bid recommendations, target promotion groups (Amazon's own harvesting), the cross-product Recommendations API launched in August 2026, and the Partner Opportunities catalog;
+- campaign types and settings where Amazon decides on the advertiser's behalf: automatic targeting, keyword groups, Sponsored Brands theme targeting and optimization rules, Sponsored Display's optimization objective, the v1 campaign model's `autoCreateTargets`, `autoManageCampaign`, automatic budget allocation and automatic global scaling, Full-Funnel Campaigns and DVA+;
+- creative automation: Creative Agent, Video Generator, image generation, auto-assembled videos (opted in by default), Sponsored Brands headline optimization, AI-powered prompt ads and the AI shopping surfaces (Alexa for Shopping, into which Rufus was merged in May 2026, and Alexa+);
+- Amazon Ads Agent (from 2026-09-29 the name of the unified ads platform, with a conversational experience in closed beta inside it);
+- Seller Central's Seller Assistant (formerly Project Amelia), which drafts Sponsored Products and Sponsored Brands campaigns, and the SP auto-optimization announced at Accelerate 2026;
+- how Amazon measured the figures it publishes for its own automation;
+- the Amazon Ads MCP server: its tools, its Lite form, its skills library and what that library says about Amazon's agent-facing method;
+- Amazon Marketing Stream as the data feed automation runs on (hourly and 15-minute data, AWS destinations);
+- the third-party tools: Helium 10 Ads (formerly Adtomic), Perpetua, Teikametrics, Scale Insights, m19, Quartile, Intentwise, Ad Badger, BidX, Adbrew, Pacvue and others, with their stated method and, as context only, their pricing model;
+- and, last, what an agent holding `cx-amazon-ads`'s open door can already do that these tools do.
+
+Why a senior Amazon advertising lead cares. Almost every lever a seller pulls has an Amazon-run twin: a bid the seller sets is moved by Amazon's bidding strategy, a budget the seller sets is raised by a budget rule or borrowed from the portfolio, a target the seller does not add is added by an auto campaign, a keyword group, a Sponsored Brands optimization rule or `autoCreateTargets`, and creative the seller never uploaded (auto-assembled videos, prompts) runs on the seller's campaigns unless the seller opts out. A lead reading an account first asks which of these are switched on, because each one changes what a report means and what a manual change will do. Third-party tools stack a second layer of automation on top (bid algorithms, harvest rules, dayparting), which writes through the same Ads API that `cx raw amazon-ads` reaches. Several of these tools, and Amazon itself, now sell an MCP connection or "skills" for Claude and other agents, which is the exact form the ads playbook takes.
+
+Other files own the levers in depth: `areas/bids-and-placements.md` (strategies, adjustment arithmetic, rule-based bidding's 5× ceiling), `areas/budgets-and-pacing.md` (budget rules as a pacing tool), `areas/search-terms-and-negatives.md` (harvesting method), `areas/keyword-and-product-targeting.md` (auto targeting's four expressions, keyword recommendations), `areas/campaign-structure.md`, `beyond/ad-products-and-eligibility.md` (Full-Funnel, DVA+, prompts as ad products), `method/ads-api-surface.md` (the endpoint inventory), `method/reporting.md` (Reporting v1 and Stream as data surfaces), `method/change-measurement.md` (why Amazon's automation confounds a before/after reading). This file states each automation once, at the level of "what Amazon decides, on what input, with what switch", and points to them.
+
+## Amazon's own rules and guidance
+
+RULE marks something Amazon enforces: an API shape, an enum, a limit, an eligibility condition, a deprecation. GUIDANCE marks a recommendation. Quotes are verbatim where they carry weight.
+
+### A. The map: what Amazon decides for an advertiser, and the switch for each
+
+[seat] The table gathers, in one place, every automation the pages below document, with the field or setting that turns it on or off. Each row is sourced in the lettered subsections that follow; the table itself is the seat's assembly.
+
+| Automation | Ad product | Who turns it on | The switch Amazon documents | Section |
+|---|---|---|---|---|
+| Dynamic bids down only | SP | default for a v3 campaign created without `dynamicBidding` | `LEGACY_FOR_SALES` (v3) / `SALES_DOWN_ONLY` (v1) | B |
+| Dynamic bids up and down | SP, SB | advertiser | `AUTO_FOR_SALES` (v3) / `SALES_UP_AND_DOWN` (v1) | B |
+| Rule-based bidding (ROAS guardrail) | SP | advertiser | `/sp/rules/campaignOptimization`; `RULE_BASED` | B |
+| Schedule bid rules | SP | advertiser | `/sp/rules/optimization` (`ruleCategory` `BID`, `ruleSubCategory` `SCHEDULE`) | B |
+| Performance optimization rules (v2 model) | SP | advertiser | `/sp/rules/optimization`, media type `...spoptimizationrules.v2+json` | B |
+| KPI-target, new-to-brand and spend-in-full strategies | v1 model | advertiser | `PRIORITIZE_KPI_TARGET`, `NEW_TO_BRAND`, `SPEND_BUDGET_IN_FULL` | B |
+| Schedule and performance budget rules | SP, SB, SD | advertiser | `/sp|sb|sd/budgetRules` plus association | C |
+| Unspent budget shared inside a portfolio | portfolios | advertiser | `budgetControls.campaignUnspentBudgetSharing` | C |
+| Automatic targeting (four expressions) | SP | advertiser (auto campaign) | auto targeting type; v1 `THEME` targets | E |
+| Keyword groups (beta, US) | SP | advertiser | `GROUP` match type; `/sp/targeting/recommendations/keywordGroups` | E |
+| Theme targeting and `smartDefault` | SB | advertiser, or Amazon when a goal is chosen | `smartDefault: ["TARGETING"]`, `/sb/themes` | E |
+| SB optimization rules (cost control; may add targets) | SB | advertiser | SB optimization rules (beta) | E |
+| Amazon-created targets | v1 model | advertiser | `autoCreateTargets: true` | E |
+| Amazon-managed campaign lifecycle | v1 model | advertiser | `autoManageCampaign: true` | E |
+| Budget moved across ad groups or marketplaces | v1 model | advertiser | `budgetAllocation: AUTO`, `marketplaceBudgetAllocation: AUTO` | E |
+| Global campaign scaled to new marketplaces | v1 model | advertiser | `AutoScaleGlobalCampaignSetting: AUTO` | E |
+| Automatic global ads free trial | several | Amazon (trial), advertiser (enrol) | read through `AUTOMATIC_GLOBAL_ADS_ENROLLMENT` | D |
+| Auto-assembled videos | SP (US) | Amazon, opted in by default | `optOuts.videoExtension: ["AUTO_ASSEMBLED"]` | F |
+| Headline optimization (DCO) | SB | advertiser consent | `creativePropertiesToOptimize: ["HEADLINE"]` | F |
+| Prompt ads | SP, SB | Amazon ("zero additional setup") | paused per prompt in the console | F |
+| Ads in Alexa for Shopping (Rufus merged into it, May 2026) and Alexa+ | SP, SB, sponsored ads | Amazon ("automatically eligible", "automatically enrolled") | none documented | F |
+| Full-Funnel Campaigns, DVA+ | cross-format | advertiser | campaign type in Amazon Ads Agent | E |
+| Auto-optimization: adds keywords and product targets, pauses keywords, under a seller-set maximum bid (US, January 2027 per Amazon Ads) | SP | advertiser ("a single click in your campaign settings") | console setting; no API field found | L |
+| Campaigns drafted by Seller Assistant from listing and inventory readiness | SP, SB | seller approves the draft | Seller Central; no API | L |
+| Seller Assistant workflows (pricing, restock, monitoring) | not ads | seller chooses "surfaces recommendations or takes actions" | Seller Central; no API | L |
+
+### B. Bidding automation
+
+The full treatment is `areas/bids-and-placements.md`; the lines here are what this file needs to describe Amazon's automation as a whole.
+
+- RULE: Sponsored Products bidding strategies in the v3 model. `LEGACY_FOR_SALES` is "Dynamic bids - down only": "Lowers your bids in real time when your ad may be less likely to convert to a sale. Campaigns created before the release of the bidding controls feature used this setting by default." `AUTO_FOR_SALES` is "Dynamic bids - up and down": "Increases or decreases your bids in real time by a maximum of 100%." `MANUAL` is "Fixed bid": "Uses your exact bid and any placement adjustments you set, and is not subject to dynamic bidding." The `BiddingStrategy` enum also carries `RULE_BASED`. [official] (Sponsored Products 3.0 OpenAPI, schemas `BiddingStrategy` and `BiddingStrategyRecommendation`, https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/SponsoredProducts_prod_3p.json, read 2026-11-09)
+- RULE: the v1 model's `BidStrategy` enum: `MANUAL`, `NEW_TO_BRAND` ("Optimizes bidding to maximize new-to-brand customer acquisitions"), `PRIORITIZE_KPI_TARGET` ("Optimizes bidding to achieve the KPI target specified"), `RULE_BASED` ("Applies bidding rules defined by the advertiser"), `SALES_DOWN_ONLY` ("Bids will never increase beyond your set bid"), `SALES_UP_AND_DOWN` ("by a maximum of 100%"), `SPEND_BUDGET_IN_FULL` ("Prioritize spending full budget, while maximizing performance") and `USE_CAMPAIGN_STRATEGY` ("Inherit the bid strategy from the parent campaign"). A campaign's `goalSettings` carries `goal` (`AWARENESS`, `CONSIDERATION`, `CONVERSIONS`, `ENGAGEMENT`), `kpi` and `kpiValue` ("The value of the KPI that the campaign is working to optimize"). Which strategies each ad product accepts is in the per-product view of the reference, not in the enum. [official] (Amazon Ads API v1 merged OpenAPI, schemas `BidStrategy`, `GoalSettings`, `Goal`, `KPI`, https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/AmazonAdsAPIALLMerged_prod_3p.json, read 2026-11-09)
+- RULE: how far up and down goes. "When you select dynamic bids - up and down, we will raise or lower your bid based on performance ... We will increase or decrease your bids by up to 100% for all placements based on performance." "Ensure you are prepared to potentially spend double the amount of the bid you input." [official] (*Guide to dynamic bidding - up and down with Sponsored Products*, Amazon Ads library, https://advertising.amazon.com/library/guides/dynamic-bidding-sponsored-products, read 2026-11-09)
+  - RULE, the dissenting Amazon page: *An author's guide to Sponsored Products* (library guide, https://advertising.amazon.com/library/guides/authors-guide-to-sponsored-products, re-opened 2026-11-09 in the gap fill) still reads "Bids will be increased by up to 100% for placements at the top of the first page of shopping results and by up to 50% for all other placements." [official]
+  - [seat] So four current Amazon sources (this library guide, which says "Currently, we can increase this bid by up to 100%, so that would be a maximum of $2.00 for all placements opportunities"; the v3 model; the v1 model; and the help page *Bidding strategies for Sponsored Products*, GCU2BUWJH2W3A8Z7, https://advertising.amazon.com/help/GCU2BUWJH2W3A8Z7: "We may increase or decrease your bids by up to 100% for all placements based on performance") say 100% for all placements, and one Amazon page, written for KDP authors, says 100% top of search and 50% elsewhere; several practitioners (BidX, Perpetua, Trellis) still use 100/50. No Amazon page dates a change from one to the other, so "the 50% is the older rule" is a reading (this seat's, PPC Land's), not Amazon's statement. `areas/bids-and-placements.md` owns this fact (P-bids-and-placements-8), and this file defers to it.
+- GUIDANCE: the same guide lists when to use up and down ("Your key objective is sales", winning top-of-page placements, "a high-performing campaign", "ASINs with excess inventory", "deal ASINs"), says "You should review your bids often, about once every two weeks", and "We recommend that you limit your changes while testing strategies, so that you can attribute the difference in performance to a specific change." [official] (same page)
+- RULE: rule-based bidding. "For any existing campaign running for at least 10 days, you can apply a rule with a guardrail of ROAS and Amazon Ads may then adjust your base bids up and down to increase conversions while maintaining your guardrails. Amazon Ads does not guarantee the ROAS guardrail. However, if your campaign is not meeting the guardrail and the return on ad spend (ROAS) of the campaign for the last 20 days drops to less than 30% of your ROAS over a 20-to-40-day period, we will disable the rule on your behalf and will enable your previous bidding strategy and targeting level bids." Eligibility: "Your campaign must be running for at least 10 days prior to setting a rule and have a minimum of 10 conversions in the last 30 days", any targeting type, and a minimum daily budget by marketplace (US, UK, DE, FR, IT, ES, CA, AU, AE, NL, BE, SG 10; IN 300; JP 600; MX 50; BR 20; EG 50; SA 20; SE 50; PL 30; TR 50, in local currency) *(for human readers only)*. "The guardrail is applied at the campaign level and the ROAS guardrail is based on the previous 21 days." When Amazon disables a rule, "you will receive an additional recommendation from AMR with an updated suggestion for your ROAS guardrail." [official] (*Sponsored Products rule-based bidding overview*, Ads API guide, https://advertising.amazon.com/API/docs/en-us/guides/rules/bidding-rules/overview, markdown at https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/rules/bidding-rules/overview.md, read 2026-11-09)
+- GUIDANCE: "We suggest that you use the provided recommendation to set your ROAS guardrail. This recommendation is based on your historical campaign information and a group of similar advertisers. Additionally, if you set your ROAS guardrails too aggressively, you may see a decrease in the number of conversions compared to your historical performance." [official] (same page)
+- RULE: the rule-based bidding API. `POST /sp/rules/campaignOptimization` creates a rule (`recurrence` `DAILY` only; `ruleAction` `ADOPT` only; `ruleType` `BID`; with `metricName` `ROAS` the operators are `GREATER_THAN` or `EQUAL_TO`); `GET`, `PUT`, `DELETE` manage it; `POST /sp/rules/campaignOptimization/state` reads per-campaign state ("We recommend that you execute this query a minimum of once per day"); `POST /sp/rules/campaignOptimization/eligibility` returns, for an eligible campaign, "a ROAS and bid guardrail guidelines ... based on the historical campaign performance of advertisers similar to you". `ruleStatus` is `ACTIVE` or `ARCHIVED`. The create request takes at most 20 `campaignIds`; the stored rule lists up to 100. [official] (*Getting started with the Sponsored Products rule-based bidding API*, https://advertising.amazon.com/API/docs/en-us/guides/rules/bidding-rules/getting-started; SP 3.0 OpenAPI, `CreateSPCampaignOptimizationRulesRequest`, `CampaignOptimizationRule`; read 2026-11-09)
+- RULE: schedule bid rules. They "allow advertisers to automatically increase bids at specific times of the day, days of the week, date ranges, or during high-traffic events like Black Friday." A rule is created with `POST /sp/rules/optimization` and attached with `POST /sp/campaigns/{campaignId}/optimizationRules` (header `application/vnd.spoptimizationrules.v1+json`); `POST /sp/rules/optimization/search` finds rules (the SP 3.0 model has no `GET` on a campaign's optimization rules; the association path is `POST` only). "Whenever a new schedule bid rule is created or an existing rule is updated, the changes will take effect within 24 hours. A maximum of 20 schedule bid rules can be specified for a campaign. Schedule bid rules are applied on top of 'Placement adjustment' and 'Bidding strategies'." Increases from rules whose conditions hold are added: a 10% daily rule, a 20% Black Friday rule and a 20% 5–9 PM rule give 30% all Black Friday and 50% from 5 to 9 PM. Event ids come from `POST /sp/v1/events`. [official] (*Schedule Bid Rules*, Ads API guide, https://advertising.amazon.com/API/docs/en-us/guides/rules/bidding-rules/scheduled-bid-rules, read 2026-11-09)
+- RULE: the shape of a schedule rule's action. `actionType` is `ADOPT` only; `actionOperator` is `INCREMENT` only; `actionUnit` is `PERCENT`; `value` is "An integer between 1 & 100, representing the percent increase on base bid". `timesOfDay` takes at most one window with `HH:00` start and end; `recurrence.type` is `DAILY` or `WEEKLY`; `status` is `ENABLED`, `ENDED`, `PAUSED` or `SCHEDULED` ("Only ENABLED and PAUSED are accepted in requests"). The association request takes at most 25 rule ids, and the schema says "Maximum 100 rules can be associated to each campaign". [official] (SP 3.0 OpenAPI, `OptimizationRulesAPISwagger*` schemas, read 2026-11-09)
+  - [seat] The guide's "maximum of 20 schedule bid rules" and the schema's "Maximum 100 rules ... to each campaign" are different counts (schedule bid rules vs all optimization rules); the model reads the live error when a write is refused.
+- RULE: a newer optimization-rule model with performance conditions. The same endpoint accepts `application/vnd.spoptimizationrules.v2+json`. In it, `ruleCategory` is `BID`, `BUDGET` or `TARGETING`; `ruleSubCategory` is `PERFORMANCE` or `SCHEDULE`; the required fields are `ruleCategory`, `ruleName`, `ruleSubCategory` and `status`, and `conditions`, `targeting` and `action` are optional; when present, `conditions` holds 1 to 10 entries, each an `attributeName` of `ACOS`, `CLICKS`, `CPC`, `CTR`, `CVR`, `IMPRESSIONS`, `ORDERS`, `ROAS`, `SALES` or `SPEND` with a value criterion (`comparisonOperator` and `value`; range criteria "NOT SUPPORTED right now"); and, when present, exactly one `targeting` block with `targetingType` `KEYWORD` or `PRODUCT`, 1 to 3 `expressionTypes` (`BROAD`, `EXACT`, `EXPANDED`, `PHRASE`) and `lookbackDays` from 3 to 65. A create request takes up to 300 rules, and its description says "Budget rules are not supported for this operation." The action remains `ADOPT` / `INCREMENT` / `PERCENT`. [official] (SP 3.0 OpenAPI, `OptimizationRulesAPISwaggerOptimizationRuleWithoutRuleIdV2`, `...RuleAttributeV2`, `...RuleCategoryV2`, `...RuleSubCategoryV2`, `...RuleTargeting`, `...CreateOptimizationRulesRequestV2`, read 2026-11-09)
+  - [seat] No Amazon guide page found on 2026-11-09 describes these v2 performance rules; the shape is readable only from the model. As modelled, every SP optimization rule (schedule or performance, v1 or v2) can only add a percentage to a bid. That is not true of Amazon's bid automation as a whole: rule-based bidding and Sponsored Brands optimization rules move bids down as well as up (next item and section E). Pausing a target on performance is what the third-party rule engines below do with direct writes.
+- RULE: the console help page on bidding rules. Rule-based bidding: "Amazon then adjusts your base bids for each ad impression, increasing or decreasing them as needed to try to achieve your performance target. Amazon does not guarantee that we'll hit the performance target that you set." "rule-based bidding can increase your bids by up to 5 times your adjusted bid amount (if any adjustments are applied to your original bid)", with a worked table ($1.00 base, 10% top-of-search adjustment: $1.10 adjusted, $5.50 max). The console eligibility reads differently from the API guide above: campaigns must meet the minimum daily budget; "You can specify a ROAS target in campaign settings. We recommend you should wait at least two weeks for the campaign to get enough data before specifying a ROAS target. If not specified, Amazon Ads will automatically determine an appropriate ROAS target to maximize sales." Schedule bid rules: "You can't apply schedule bid rules to campaigns that have Rule-based bidding active. Schedule bid rules will be on hold if you change your bidding strategy to Rule-based bidding"; they apply to "Dynamic bids - up and down", "Down only" and "Fixed"; "Schedule bid rules are add-on" (a 20% and a 10% rule give 30%), and "Schedule bid rules will be applied on top of your placement adjustment" (down only, 10% top-of-search, $1.00 bid, 20% Monday rule: $1.30 top of search and $1.20 elsewhere on Mondays). The hourly campaign report "is available for the last 30 days" *(for human readers only)*. [official] (*Bidding rules for Sponsored Products*, help GDLQ5D2BNFCAU3TW, https://advertising.amazon.com/help/GDLQ5D2BNFCAU3TW, read 2026-11-09 by the verifier from the page's embedded content on a plain GET)
+  - [seat] The API guide's "at least 10 days ... minimum of 10 conversions in the last 30 days" and the help page's optional, Amazon-set ROAS target are two Amazon pages that disagree on eligibility; the eligibility call's live answer is the reading that holds for a given campaign.
+- RULE: audience bid boosting is Amazon's segment-level automation on the bid. Sponsored Products carries three Amazon-built audiences ("Purchased brand products", "Clicked or added brand product to cart," "Affinity for similar products") and Sponsored Brands carries "New-to-brand shoppers" plus "Purchased brand product" and "Clicked or Added brand product to cart"; the ids come from `ListTargetableEntities` with `targetTypeFilter` `AUDIENCE`, used with `audienceSegmentType` `BEHAVIOR_DYNAMIC`. The Sponsored Products entry (2025-03-12) is headed "Preview" and says "This capability is in preview only. We will update our release notes once this functionality is made available"; no later SP entry in the release notes on 2026-11-09 announces it generally available. The Sponsored Brands "New-to-brand shoppers" audience went worldwide on 2025-03-24, and the two further SB audiences arrived on 2025-08-22. [official] (Ads API release notes, entries of 2025-03-12, 2025-03-24 and 2025-08-22, https://d3a0d0y2hgofx6.cloudfront.net/en-us/release-notes/index.md, read 2026-11-09)
+
+### C. Budget automation
+
+The pacing method is `areas/budgets-and-pacing.md`; these are the Amazon-run mechanisms.
+
+- RULE: budget rules exist for Sponsored Products, Sponsored Brands and Sponsored Display. "Schedule-based rules enable you to set a rule to increase daily budget based on scheduled events such as Black Friday or a custom time period." "Performance-based rules enable you to set rules to increase campaign budgets when your campaigns meet a certain performance threshold ... These metrics are calculated using data from the previous 7 days for sellers and 14 days for vendors." [official] (*Sponsored ads budget rules overview*, https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/overview, read 2026-11-09)
+- RULE: performance-rule metrics by ad type: `ACOS`, `CTR`, `CVR` (SP, SD); `IS`, top-of-search impression share, "not available as a metric for any campaign that contains a video creative" (SB); `NTB` (SB); `ROAS` (SB, SP, SD). Operators: `LESS_THAN`, `GREATER_THAN`, `EQUAL_TO`, `LESS_THAN_OR_EQUAL_TO`, `GREATER_THAN_OR_EQUAL_TO`. Endpoints: `POST /sp/budgetRules`, `POST /sb/budgetRules`, `POST /sd/budgetRules`, then `POST /{sp|sb|sd}/campaigns/{campaignId}/budgetRules` to associate. [official] (*Performance-based rules*, https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/performance-based, read 2026-11-09)
+- RULE: schedule rules take `DAILY`, `WEEKLY` or `intraDaySchedule` recurrence; "`intraDaySchedule` is currently only available for campaigns in the following marketplaces: US, CA, UK, IN, JP" *(for human readers only)*; hours-of-day rules need a minimum budget to become active. "Recommended events are only available for Sponsored Products and Sponsored Brands": `POST /sb/campaigns/budgetRules/recommendations` (and the SP equivalent `POST /sp/campaigns/budgetRules/recommendations`) returns events with `suggestedBudgetIncreasePercent`; events listed before their dates are announced get their dates filled in later automatically. [official] (*Schedule-based rules*, https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/schedule-based; SP 3.0 OpenAPI path list; read 2026-11-09)
+- RULE: evaluation. "Budget rules are evaluated on your campaigns at the start of the day to calculate campaign budget for the day." A rule is also evaluated "immediately" when created or updated, and whenever the campaign's daily budget changes. "Multiple rules can be specified for a campaign to a maximum of 250." Since February 2023, "If more than one budget rule satisfies your set conditions, all budget rules are applied and their increases are added to daily budget independently of other budget rules." Statuses: Active, On Hold (criteria not met, superseded, recurrence not met), Paused, Budget threshold not met (performance rules only), Pending Start, Expired. [official] (*Budget rule evaluation*, https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/rule-evaluation, read 2026-11-09)
+  - [seat] The page's own worked example contradicts itself: with campaign ROAS 4, it applies the 40% rule (ROAS > 3) and says the 20% rule (also ROAS > 3) "was not applied because its condition was not met". The additive rule above is the stated logic; the example is not a reliable reading of edge cases.
+- RULE: portfolio unspent-budget sharing. "when some campaigns do not use their full daily budget, that unspent campaign budget ... can automatically flow to other campaigns in your portfolio that consistently reach their daily spending limits ... high-performing campaigns [can] spend up to twice their daily budget by tapping into unspent campaign budget from previous days in the month ... while staying within your total monthly portfolio budget." Switch: `budgetControls.campaignUnspentBudgetSharing` `ENABLED` or `DISABLED`. [official] (Ads API release notes, "Automatically share unspent campaign budget with budget exhausted campaigns in a portfolio - updated", 2025-07-19, read 2026-11-09)
+- RULE: budget recommendations. `POST /sp/campaigns/budgetRecommendations` returns, per campaign, a "Recommended daily budget" ("Estimated daily budget needed to keep the campaign in budget for the full 24-hour period"), "Percent time in budget" ("value -1 means we don't have enough information to compute the campaign's percent time in budget"), and "Estimated missed impressions, clicks and sales" ranges ("estimates based on historical traffic and the campaign's past performance ... but not guaranteed"). `POST /sp/campaigns/initialBudgetRecommendation` "Creates daily budget recommendation along with benchmark metrics when creating a new campaign." `POST /sp/campaigns/budget/usage` (and SB, SD equivalents) reads budget usage. [official] (SP 3.0 OpenAPI, operation descriptions, read 2026-11-09). The Partner Opportunities catalog says these "are refreshed once per day". [official] (*Partner Opportunities - Campaign Optimization*, https://advertising.amazon.com/API/docs/en-us/guides/recommendations/partner-opportunities/catalog/campaign-optimization, read 2026-11-09)
+
+### D. Recommendations and the "apply" flows
+
+- RULE: consolidated Sponsored Products recommendations. `GET` or `POST /sp/campaign/recommendations` "Gets the top consolidated recommendations across bid, budget, targeting for SP campaigns given an advertiser profile id. The recommendations are refreshed everyday." Each campaign entry may carry `biddingStrategyRecommendation` (`action` `UPDATE`, `suggestedBiddingStrategy` `AUTO_FOR_SALES` / `LEGACY_FOR_SALES` / `MANUAL`), `budgetRecommendation` (`INCREASE` or `DECREASE`, `suggestedBudget`), up to 50 `keywordTargetingRecommendations` (`ADD`, `DECREASE`, `INCREASE`, `REMOVE`, `UPDATE`; `keywordText`, `matchType` including `GROUP`, `suggestedBid`), up to 50 `placementBiddingRecommendations` (`PLACEMENT_TOP`, `PLACEMENT_REST_OF_SEARCH`, `PLACEMENT_PRODUCT_PAGE`; `suggestedBidAdjustment`; `incrementalImpressionsLowerPercent` and `...UpperPercent`), up to 50 `targetingGroupBidRecommendations` for the four auto expressions, and `sevenDaysEstimatedOpportunities` (`estimatedIncrementalClicksLower`, `...Upper`, "if all optimizations are made"). Placement and targeting-group recommendations carry an `action` of `ADD`, `DECREASE`, `INCREASE` or `REMOVE`. That shape is the `GET` response (`application/vnd.spgetcampaignrecommendationsresponse.v1+json`). The `POST` form answers `application/vnd.spgetcampaignrecommendationsresponse.v2+json`: a list of up to 50 `recommendations`, each with `campaignId`, `recommendationType` (`BIDDING_STRATEGY`, `KEYWORD`, `KEYWORD_GROUP`, `BUDGET_STRATEGY`, `PLACEMENT_BIDDING`, `SHOPPER_COHORT`), `recommendationDetails` and `forecastEstimates`, paged by `nextToken`. Permission `advertiser_campaign_view` or `advertiser_campaign_edit`. [official] (SP 3.0 OpenAPI, tag "Consolidated Recommendations", schemas `CampaignRecommendation`, `GetCampaignRecommendationsResponseV2`, `Recommendation`, `RecommendationType`, read 2026-11-09)
+  - [seat] There is no "apply" endpoint for these: a recommendation is applied by writing the campaign, keyword or placement it names through the ordinary update operations. The one-click "apply" in the console is that write.
+- RULE: theme-based bid recommendations. `POST /sp/targets/bid/recommendations` (`application/vnd.spthemebasedbidrecommendation.v3+json`; later versions add `includeAnalysis`, "Flag to include new bid analyzer data") takes `BIDS_FOR_EXISTING_AD_GROUP` (campaign and ad group ids) or `BIDS_FOR_NEW_AD_GROUP` (up to 50 ASINs, a bidding strategy and up to 3 placement adjustments) and returns, per theme such as `CONVERSION_OPPORTUNITIES`, three suggested bids per targeting expression ("minimum, median, and maximum suggested bids") with `impactMetrics` ranges for clicks and orders. A multi-country form is `POST /sp/global/targets/bid/recommendations`. [official] (*Sponsored Products auto targeting*, https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/auto-targeting; SP 3.0 OpenAPI, `AsinsThemeBasedBidRecommendationRequestV5`, `MultiCountryThemeBasedBidRecommendationResponse`; read 2026-11-09)
+- RULE: Amazon's own harvesting, target promotion groups. "Target promotion is a performance advertising strategy that identifies targets for further refinement ... This combines the common practice of target harvesting, which identifies targets to move from broader targeting (for example, automatic targeting) to more focused targeting (for example, manual targeting with broad match), with optimized bids and budget changes." `POST /sp/targetPromotionGroups` either creates new manual keyword and product campaigns from an auto ad group (`newCampaignDetails`: `namePrefix`, dates, `dynamicBidding`, daily `budget`, `defaultBid`) or links one existing keyword ad group and one existing product ad group whose ads carry "the same Ad ASINs/SKUs combination". `POST /sp/targetPromotionGroups/recommendations` "Retrieves keyword and product targets of an auto-targeting campaign as recommendations for promoting to a manual-targeting campaign. The recommendations are based on performance heuristics of the targets", each with `recommendationReasons` such as `HISTORICAL_PERF` with `CLICKS` and `CONVERSIONS`. `POST /sp/targetPromotionGroups/targets` adds targets (up to 1,000 a call) with `expressionType` `QUERY_BROAD_MATCHES`, `QUERY_EXACT_MATCHES`, `QUERY_PHRASE_MATCHES`, `ASIN_SAME_AS` or `ASIN_EXPANDED_FROM`; `.../list` and `.../targets/list` read them back. [official] (*Target Promotion*, https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/target-promotion; SP 3.0 OpenAPI, tag `TargetPromotionGroups`; read 2026-11-09)
+  - RULE: the Partner Opportunities catalog describes the same recommendations as "Recommendations for Automatic Targeting (AT) to Manual Targeting (MT) ad target conversions ... along with the historical clicks, conversions, conversion rate, and ROAS for the recommended target from the past 64 days." [official] (*Partner Opportunities - Campaign Optimization*, above)
+- RULE: the cross-product Recommendations API (launched 2026-08-31). The release note calls it "a single interface for retrieving recommendations across ad products"; the overview says "a single interface for retrieving recommendations across Amazon advertiser campaigns" and "Each recommendation includes a unique identifier for tracking, a recommended action, and insights related to why the customer is receiving the recommendation." `GET /adsApi/v1/list/recommendations` ("pre-computed", header `Amazon-Ads-AccountId` with a global account id `amzn1.ads-account.g.…` or a DSP account id; without it the gateway answers `401`) and `POST /adsApi/v1/query/recommendations` (body `accessRequestedAccounts` with exactly one `advertiserAccountId` and `marketplacesFilter`; optional `languageFilter`, `adProductFilter`, `recommendationTypeFilter`, `recommendationTypeDetailsFilter`; `maxResults` default and maximum 300). Without a type filter the query runs in "Discovery mode", "the primary interface for automated systems that evaluate multiple recommendation types". A recommendation has `recommendationType`, `recommendationInsight.insightTexts`, `state` (`ACTIVE`, `INACTIVE`, `EXPIRED`, `APPLIED`, `DEFERRED`, `DISMISSED`), `potentialImpacts`, `expiryDateTime`, and `recommendedObjects[]` with `currentObjectSettings`, `recommendedObjectSettings` and `recommendedAction` (`linkText`, `method` `GET`/`POST`/`PATCH`/`DELETE`, `href`). "Design your application to tolerate `recommendationType` values it does not recognize". "Pagination is not currently supported on either endpoint. Results are capped at `maxResults` (maximum `300`)"; rate limits are "`5` requests per second, with a burst allowance of `10` requests" per advertising account *(for human readers only)*. [official] (*Recommendations API overview*, https://advertising.amazon.com/API/docs/en-us/guides/recommendations/recommendations-api/overview; *How to call the API*, https://advertising.amazon.com/API/docs/en-us/guides/recommendations/recommendations-api/how-to; release notes 2026-08-31; read 2026-11-09)
+- RULE: as of 2026-11-09 the Recommendations API serves four types: `REMOTE_CAMPAIGN_PAUSED` (SP; "a campaign has been paused by the system rather than by the advertiser, for example in a remote marketplace"), `AUTOMATIC_GLOBAL_ADS_ENROLLMENT` (SP, SB, SD, ST, DSP, Full Funnel; `enrollmentStatus` `ELIGIBLE_FOR_FREE_TRIAL_ENROLLMENT`, `FREE_TRIAL_STARTED`, `FREE_TRIAL_EXPIRING`, `FREE_TRIAL_ENDED`, `ELIGIBLE_FOR_ENROLLMENT`, with `trialMetrics`), `BRAND_PRODUCT_SEGMENT` and `BRAND_PRODUCT_KEYWORD` (SB; scoped mode only, with a required detail payload; the keyword type returns `keywordTarget`, `recommendedBid` with `suggested`/`lower`/`upper`, and `keywordTheme` such as `Generic`, `Branded`, `Seasonal`). "Recommendation types beyond the table above are not currently available through these endpoints; requesting them returns `400 BadRequest` with the error code `INVALID_RECOMMENDATION_TYPE`." [official] (*Recommendation types*, https://advertising.amazon.com/API/docs/en-us/guides/recommendations/recommendations-api/recommendation-types, read 2026-11-09)
+  - RULE: the example response for automatic global ads reads: "We expanded some of your eligible Sponsored Products campaigns to additional countries at no cost. Your campaigns will pause when the trial ends in 18 days until you choose to enroll." and "You'll only be charged for ad spend in these countries if you choose to enroll." [official] (*How to call the API*, example response) — [seat] that is an Amazon-initiated expansion of a seller's campaigns into other countries, readable only through this recommendation type.
+  - [seat] The merged v1 model lists `GET /adsApi/v1/recommendations` (`ListRecommendation`) and SB-specific `POST /adsApi/v1/create/recommendations/sb` and `POST /adsApi/v1/query/recommendationTypes/sb`, while the guide writes `GET /adsApi/v1/list/recommendations`. The path is read from the reference at run time.
+- RULE: what the deprecation moves. Amazon Marketing Stream Version 1 subscriptions, "including reporting data, messaging data, and recommendation data", stop on June 30, 2027 with v3 and Sponsored Ads v2 reporting; "For messaging data (entity state changes and budget consumption notifications), use the Campaign Management API. For recommendation data, use the Recommendations API. These features are not carried forward in Stream v2." [official] (*Deprecations*, "September 2026 - Deprecation of Amazon Marketing Stream Version 1, v3 reporting, and Sponsored Ads v2 reporting", https://advertising.amazon.com/API/docs/en-us/release-notes/deprecations, read 2026-11-09)
+- RULE: two recommendation-adjacent Sponsored Brands reads are ending: the SB benchmark APIs shut off on 2026-12-17, and the SB v2 `searchTermImpressionShare` and `searchTermImpressionRank` return NULL from 2026-11-10, replaced by `metric.impressionShare` and `metric.impressionShareRank`. [official] (*Deprecations*, read 2026-11-09)
+- RULE: the Partner Opportunities API is a partner surface ("Help your advertisers ..."): it lists opportunity types such as target promotion groups, budget-exhaustion risk for SP, SB and SD, unadvertised new ASINs, "Create Sponsored Products campaign with presets" ("Advertisers who created a new Sponsored Products campaign using preset recommendations saw 29% more ad-attributed sales on average after one month, compared to similar \* advertisers", footnoted "Amazon internal data, WW, September 2022 – August 2023" *(for human readers only)*), products with rising ratings, and products with upcoming coupons. [official] (*Partner Opportunities - Campaign Optimization*, above)
+- RULE: Sponsored Brands recommendation helpers: the optimization-rule recommendation takes a cost-control metric type and up to ten landing-page ASINs or URLs and returns "both the recommended value and minimum value" for an optimization rule (2024-10-18); the release note writes the path `/sb/recommendation/optimization`, while the SB 3.0 model and the SB optimization rules guide write `POST /sb/recommendations/optimization` (media type `application/vnd.sboptimizationrecommendationresource.v4+json`). `POST /sb/campaigns/insights` returns, for keywords, search-term impression share and rank and flags for low traffic and low winning probability (2023-03-07). [official] (Ads API release notes; Sponsored Brands 3.0 OpenAPI path list, https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/SponsoredBrands_prod_3p.json; *Managing Sponsored Brands optimization rules*; read 2026-11-09)
+- RULE: other Sponsored Products recommendation reads the API offers: `POST /sp/targets/keywords/recommendations` (ranked keywords), `POST /sp/global/targets/keywords/recommendations/list`, `POST /sp/targets/products/recommendations`, `POST /sp/targets/categories/recommendations`, `GET /sp/negativeTargets/brands/recommendations`, `POST /sp/targeting/recommendations/keywordGroups`. [official] (SP 3.0 OpenAPI path list, read 2026-11-09). Their method is `areas/keyword-and-product-targeting.md`'s and `areas/search-terms-and-negatives.md`'s.
+
+### E. Campaign types and settings where Amazon decides
+
+- RULE: automatic targeting. Four expressions are created for every auto ad group; in v1 they are `THEME` targets (for example `themeTarget.matchType` `KEYWORDS_CLOSE_MATCH`); "If you do not set up any auto targeting clauses, your ad group uses the default bid you assigned during ad group creation for all targeting." [official] (*Sponsored Products auto targeting*, above). [seat] The guide's "Get target IDs" step shows `POST /adsApi/v1/create/targets` with a filter body, which reads as a documentation slip for the query operation; the model reads the reference.
+- RULE: keyword groups. "Keyword Groups targeting automatically identifies and targets highly relevant keyword targets for ads ... based on specified product `ASIN`s and continuously refreshes them over time"; "currently in beta and only available for US marketplace"; it can sit with keywords in one ad group but not with product targets. The recommendation endpoint describes it as dynamically "updating the keywords within a group through the campaign lifecycle", with performance "in the search terms report". [official] (*Sponsored Product keyword groups targeting*, https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/keyword-group-targeting; SP 3.0 OpenAPI `getKeywordGroupRecommendations`; read 2026-11-09)
+- RULE: Sponsored Brands theme targeting and smart defaults. Themes are `KEYWORDS_RELATED_TO_YOUR_BRAND` and `KEYWORDS_RELATED_TO_YOUR_LANDING_PAGES` ("Additional theme types may be added in the future"); "If you create a campaign with the `smartDefault` parameter with `["TARGETING"]` value, the theme-based targeting is automatically created based on the `goal` type"; "The bid is only mutable when the corresponding campaign of the keyword does not have any enabled optimization rule." [official] (*Sponsored Brands theme-based targeting*, https://advertising.amazon.com/API/docs/en-us/guides/sponsored-brands/targeting/theme-based-targeting, read 2026-11-09)
+- RULE: Sponsored Brands optimization rules (beta). For `PAGE_VISIT` goals a CPC cost rule lets "machine learning models ... define the bids while adjusting base bids up and down"; with a campaign goal, "if costs are too high, bids are reduced, and if they are below the target, they are increased"; "the system may add targets while your campaign is running to try to stay at or below a cost per order value you have specified"; "if the campaign is not meeting the cost target over a 21-day period (excluding special days), we recommend adjusting cost target or budget". When a goal is selected "`smartDefault` will be automatically added with `TARGETING`", and "in the future, `OPTIMIZATION` can be automatically added"; `MANUAL` keeps full control. [official] (*Managing Sponsored Brands optimization rules*, https://advertising.amazon.com/API/docs/en-us/guides/sponsored-brands/campaigns/managing-optimization-rules, read 2026-11-09)
+- RULE: Sponsored Display's optimization objective. On 2025-04-04 Amazon moved "Reach and Page Visit campaigns to align with the optimization objective of Sponsored Display Conversion campaigns", so "advertisers who have active Reach and Page Visit campaigns are likely to observe increases in performance metrics". [official] (release notes, read 2026-11-09). Sponsored Display is now sold inside DVA+ (section G and `beyond/ad-products-and-eligibility.md`).
+- RULE: the v1 campaign model's automation flags. `autoCreateTargets`: "Gives Amazon permission to automatically create targets associated with the campaign based on the products being advertised." `autoManageCampaign`: "Flag that allows Amazon to manage the lifecycle of your Campaign." `budgetAllocation` `AUTO`: "Automatically allocate budget to better performing ad groups based on the selected goal KPI." `marketplaceBudgetAllocation` `AUTO`: "Auto distribute global budget to marketplaces in global campaign." `AutoScaleGlobalCampaignSetting` `AUTO`: "Auto scale global campaign to new marketplaces." `offAmazonBudgetControlStrategy`: `MAXIMIZE_REACH` or `MINIMIZE_SPEND`. [official] (Amazon Ads API v1 merged OpenAPI, schemas `AutoCreationSettings`, `BudgetAllocation`, `MarketplaceBudgetAllocation`, `AutoScaleGlobalCampaignSetting`, `OffAmazonBudgetControlStrategy`, read 2026-11-09)
+  - [seat] These are the fields a lead reads first in a v1 campaign query: each one hands a decision to Amazon, and none of them shows up as a change in a change log the seller made.
+- RULE: Full-Funnel Campaigns. "Advertisers provide their budget, products, and creative, and Amazon's AI handles the rest on their behalf—planning, executing, and continuously optimizing campaigns." It joins sponsored ads, display, video and streaming TV in one campaign; reporting adds new-to-brand and long-term sales; "Full-Funnel Campaigns is now available to all advertisers in the United States." [official] (*Introducing Amazon Ads Agent*, news, 2026-09-29, https://advertising.amazon.com/library/news/amazon-ads-agent; HTML title "Introducing Amazon Ads Agent, your AI-powered ad platform", on-page headline "Amazon Ads introduces Full-Funnel Campaigns, DVA+, and enhanced chat experience in AI-powered ad platform, Amazon Ads Agent"; read 2026-11-09). Amazon's result figure is not on that news page. It is on the unBoxed 2026 page ("Beta testers saw 67% higher long-term return on ad spend and 29% lower cost to acquire new-to-brand customers") and on the launch page ("67% higher Long-Term ROAS* compared to the same formats configured independently", footnoted "Results based on Full-Funnel Campaigns against the blended average of other campaign performance across beta advertisers. Amazon internal (June – August 2026). Early beta advertiser data; individual results may vary.") *(for human readers only; how it was measured is in section M)*. [official] (*The 11 biggest takeaways from unBoxed 2026*, https://advertising.amazon.com/library/news/unboxed-2026-news-announcements; *Grow total sales with Full-Funnel Campaigns*, what's new, https://advertising.amazon.com/resources/whats-new/full-funnel-campaigns-from-discovery-to-sales; read 2026-11-09 in the gap fill)
+- RULE: DVA+ "combin[es] what was previously managed through separate campaign types—including Sponsored Display, Sponsored TV, and programmatic campaigns ... Advertisers pick a goal—such as driving sales, increasing traffic, or building awareness—add their products and creative, set a budget, and launch. Amazon's AI handles the rest." "Advanced settings" keep targeting, deals, supply, frequency caps and bidding strategies; "DVA+ will begin rolling out to advertisers beginning in late October." [official] (*Introducing Amazon Ads Agent*, above)
+- RULE: Review Requests (beta, US) is a new campaign type billed "only when a shopper leaves a rating or review", for "products with fewer than 1,000 total reviews"; review metrics are in the console, "review-specific metrics in the reporting API will follow". The release-notes page dates this entry November 16, 2026, a week after this reading. [official] (release notes, read 2026-11-09). Its eligibility and policy are `beyond/ad-products-and-eligibility.md`'s.
+
+### F. Creative automation and the AI shopping surfaces
+
+- RULE: Creative Agent. At unBoxed 2025 Amazon made the September creative tool "available in the unified Amazon Ads console" and able to "create streaming TV ads"; it combines "Amazon's signals with conversational guidance to produce high-quality advertising content quickly and at no cost", across display, video and audio. Availability: North America, United States; who can use it: vendors, registered sellers, managed-service advertisers and self-service advertisers; access in the advertising console and Creative Studio. [official] (*Transform campaign and asset development with Creative Agent*, what's new, 2025-11-11, https://advertising.amazon.com/resources/whats-new/unboxed-2025-creative-agent, read 2026-11-09)
+- RULE: the agentic creative tool's method. In Creative Studio a "conversational, AI-powered creative partner ... conducts product and audience research, brainstorms ideas, develops creative concepts in a storyboard format, and produces compelling video and display ads"; assets are saved to the Creative Asset library to "run ads across Sponsored Brands, Sponsored Display, Amazon DSP, and Brand Stores"; it "is currently in beta" and uses foundation models on Amazon Bedrock "including Amazon Nova and Anthropic Claude". Jay Richman, Vice President, Amazon Ads, is quoted: "With our new AI creative partner in Creative Studio, businesses can explore bold ideas, collaborate in real time, and bring professional ads to life—always on, at no extra cost." [official] (*Amazon Ads launches new agentic AI tool that creates professional-quality ads*, news, 2025-09-17, https://advertising.amazon.com/library/news/amazon-ads-agentic-ai-creative-tool, read 2026-11-09)
+- RULE: creative tools inside Ads Agent's chat (unBoxed 2026): "up to 71% more product coverage and up to 15% sales growth" from advertisers using them *(for human readers only)*. [official] (*The 11 biggest takeaways from unBoxed 2026*, above)
+- RULE: Video Generator for Sponsored Brands video. Open beta, expanded on 2025-11-11 from the US to Canada, India, Mexico, France, Germany, Italy, Spain and the UK; it "transforms a single product image or video and ASIN into high-quality multi-scene videos"; "Advertisers can create up to six 15-second professional video ads within minutes ... at no additional cost", drawing on "product detail pages, listings, A+ content"; access through Creative Studio in the Amazon Ads Console. [official] (*Video Generator is now available for Sponsored Brands video campaigns in nine marketplaces*, what's new, 2025-11-11, https://advertising.amazon.com/resources/whats-new/video-generator-for-sponsored-brands-video-campaigns, read 2026-11-09)
+- RULE: auto-assembled videos for Sponsored Products (US). "Auto-assembled videos are assembled from existing uploaded video clips. No new content is generated using AI." They "serve on eligible video placements. On all other placements, the same ad continues to render as a standard image." "All campaigns are opted in to auto-assembled videos by default." Opting out is `optOuts.videoExtension: ["AUTO_ASSEMBLED"]` on `POST /adsApi/v1/update/campaigns` (an empty list opts back in); "Opting out is forward-looking. It prevents new auto-assembled videos from being created for the campaign. Existing auto-assembled videos that are already delivering are not automatically paused — use `POST /adsApi/v1/update/adExtensions` to pause them." `videoType` reads `SPOTLIGHT` ("uploaded by the advertiser") or `AUTO_ASSEMBLED` ("generated by Amazon on behalf of the advertiser"). The release note of 2026-08-09 announces the documentation ("Documentation is now available"), not the feature's start date. [official] (*Auto-assembled videos*, Ads API guide, https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/auto-assembled-videos; release notes 2026-08-09; read 2026-11-09)
+  - [seat] This is the one Amazon automation on this list that is on unless the seller acts, and the opt-out is a v1 write the open door can make.
+- RULE: Sponsored Brands headline optimization. "Dynamic Creative Optimization (DCO)" tests "multiple headlines ... and automatically serve[s] the best performing one"; consent is `creativePropertiesToOptimize: ["HEADLINE"]` "on Product Collection or Store Spotlight creatives"; Amazon quotes "up to 1.6% `CTR` lift" *(for human readers only)*. [official] (release notes, 2025-04-17, read 2026-11-09). The Product Collection ad type is itself deprecated: from August 2026 new Product Collection creates become Manual Collection ads, and they return an error after January 2027. [official] (*Deprecations*, "June 2026 - Deprecation of Sponsored Brands Product Collection Ad Type", read 2026-11-09)
+- RULE: prompt ads. The `spPromptAdExtension` and `sbPromptAdExtension` reports cover "AI-powered prompt ads", "a new ad format that integrates into your existing Sponsored Products and Sponsored Brands campaigns with zero additional setup required"; "Prompts with clicks will show in your existing Sponsored Products or Sponsored Brands reporting, and you can pause individual prompts through the Amazon Ads console." [official] (*Prompt Ad Extension reports*, Ads API reporting guide, https://advertising.amazon.com/API/docs/en-us/guides/reporting/v3/report-types/prompt-ad-extension, read 2026-11-09). The product is `beyond/ad-products-and-eligibility.md`'s.
+- RULE: AI shopping placements Amazon adds on its own, in date order. (1) 2024-09-05, US: "your ads may appear in Rufus-related placements ... Your campaign reports will not include Rufus metrics." (2) 2025-07-15: "Sponsored products campaigns will now be automatically integrated into shopping conversations when customers chat with Alexa+ about products"; "Performance metrics will be consolidated within existing reporting dashboards"; "only available on the Echo Show" at that time. [official] (release notes, read 2026-11-09). (3) May 2026: Amazon "brought together ... Rufus ... and Alexa+ ... into a single, unified experience: Alexa for Shopping", used in the main Amazon search bar and on product pages. "Existing sponsored ads campaigns are automatically eligible to serve in Alexa for Shopping ... without additional campaign setup." Sponsored Products and Sponsored Brands prompts "appear in shopping results and product pages, opening product-related conversations with Alexa for Shopping"; "Existing sponsored ads campaigns are automatically eligible to be optimized as a prompt." On Alexa+ devices, "Alexa+ Conversational Ads ... Advertisers with sponsored ads campaigns are automatically enrolled." Amazon's figures: "Nearly 20% of shoppers who interact with a prompt continue the conversation about that brand, and adding prompts to a Sponsored Brands ad drives a 6% increase in conversions" *(for human readers only)*. [official] (*What agentic shopping means for advertising* (headline "Agentic shopping is here. What does that mean for advertisers?"), news, 2026-06-11, https://advertising.amazon.com/library/news/agentic-shopping-advertising, read 2026-11-09 by the verifier). unBoxed 2026 added "Branded Conversations", brand-supplied information that "Alexa for Shopping can provide" when a shopper clicks a prompt. [official] (*The 11 biggest takeaways from unBoxed 2026*, read 2026-11-09)
+  - RULE, the merger at its source (re-opened in the gap fill): "Amazon brings together Rufus and Alexa+ to create 'Alexa for Shopping' on the Amazon Shopping app and website", "available to U.S. customers on the Amazon Shopping app and website, and Echo Show devices"; Alexa+ itself goes on "across hundreds of millions of devices, Alexa.com, and the Alexa app". [official] (*Meet Alexa for Shopping, your personalized, agentic AI assistant on Amazon*, aboutamazon, `datePublished` 2026-05-13, https://www.aboutamazon.com/news/retail/alexa-for-shopping-ai-assistant, read 2026-11-09). The June 2026 Amazon Ads page keeps two ad surfaces apart: "Conversational ad types available within Alexa for Shopping" (sponsored ads and SP/SB prompts) and "Conversational ad types available with Alexa+ and Alexa-enabled devices" (Alexa+ Conversational Ads, Prime Video Sponsored Tiles). [official] (*What agentic shopping means for advertising*, re-opened 2026-11-09). Amazon Ads' own small-business page of 2026-09-22 also names it: prompts are "ads that open deeper product and brand conversations within Alexa for Shopping", and "Shoppers who click a Sponsored Products or Sponsored Brands prompt convert to a sale 48% more often and spend 21% more on average than those who don't" *(for human readers only; a comparison of shoppers who chose to click, section M)*. [official] (*Amazon Ads introduces new sponsored ads tools to help small businesses achieve their goals*, expert advice, 2026-09-22, https://advertising.amazon.com/library/expert-advice/sponsored-ads-small-business-tools, read 2026-11-09). At Accelerate 2026 Amazon said "more than 350 million customers used Alexa for Shopping in the past year" *(for human readers only)*. [official] (*At Accelerate 2026, Amazon shares new AI tools as a force multiplier for independent sellers*, https://sellingpartners.aboutamazon.com/accelerate-2026-recap, read 2026-11-09)
+  - [seat] So, as of 2026-11-09: the Rufus surface no longer exists under that name in the US (it is Alexa for Shopping, on the app, the website and Echo Show), while Alexa+ continues as the device and Alexa.com assistant with its own ad forms; outside the US Rufus keeps its name (`beyond/ad-products-and-eligibility.md` §S). "Alexa for Shopping" does appear on Amazon Ads pages (the two above). A file that treats "Rufus" as a current US surface is reading pre-May-2026 pages; a file that treats Alexa+ devices as a separate ad surface from Alexa for Shopping matches Amazon's June 2026 page. Ownership of this fact is `beyond/ad-products-and-eligibility.md`'s.
+  - [seat] No page read documents a campaign-level switch for serving in Alexa for Shopping or Alexa+; the only documented control is pausing individual prompts in the console. Whether the 2024 "reports will not include Rufus metrics" line still holds after the merger is not stated on any page read; the June 2026 news speaks only of "closed-loop measurement".
+- RULE: the AI creative solutions page groups a conversational "creative partner", one-click image generation, one-click video generation and audio ad generation, all in Creative Studio and "integrated within Amazon Ads Agent", with case studies; the video tool is listed "for Sponsored Brands video under Creative Studio" in Australia, Canada, France, Germany, India, Italy, Japan, Mexico, Spain, the United Kingdom and the United States *(for human readers only)*. [official] (*AI creative solutions* (title "Creative Solutions - Optimize your creative with AI-powered solutions"), https://advertising.amazon.com/generative-ai-ad-solutions, read 2026-11-09)
+
+### G. Amazon Ads Agent: the platform name, and the conversational experience inside it
+
+- RULE: name and scope. "Last year at unBoxed, Amazon Ads announced the unification of the demand-side platform and ads console into a single media buying experience ... Today at unBoxed 2026, we announced Amazon Ads Agent—the new name of the unified platform, now powered by agentic AI." It "brings three campaign types—sponsored ads, Full-Funnel Campaigns, and Display, Video, and Audio—together in one simple, connected, AI-first ads platform" (the last sold as DVA+). Advertisers can "use manual campaign building tools, let Amazon's AI optimize on their behalf, or employ a mix of both." [official] (*Introducing Amazon Ads Agent*, news, 2026-09-29, https://advertising.amazon.com/library/news/amazon-ads-agent, re-opened 2026-11-09 in the gap fill)
+- RULE: the help centre's "formerly known as Amazon DSP". Two help pages write the platform name where they once wrote "Amazon DSP", in passages about DSP-specific objects: "Amazon Ads Agent (formerly known as Amazon DSP) suggests a new maximum average CPM to help improve pacing and performance" (*Understand delivery alerts and recommendations*, help GR6VEAG2EQWG56RD, https://advertising.amazon.com/help/GR6VEAG2EQWG56RD) and "Instead of creating and maintaining separate accounts for Amazon Ads Agent (formerly known as Amazon DSP) per country and for sponsored ads, you can create 1 Advertising Account" (*Understand Advertiser Accounts*, help GXKXR56ZCMJ3WCS4, https://advertising.amazon.com/help/GXKXR56ZCMJ3WCS4). [official] (both read 2026-11-09 in the gap fill)
+  - [seat] How the three descriptions fit. "Amazon Ads Agent" names one thing, the unified platform (the former ads console plus the former Amazon DSP, called Campaign Manager at unBoxed 2025), from 2026-09-29. Inside it sits "the conversational experience in Amazon Ads Agent", which is the closed beta (next item). The help centre's "(formerly known as Amazon DSP)" refers to the same platform name, written into pages whose subject is the DSP side; the news page says the platform also holds sponsored ads, so "Ads Agent = the former DSP" is too narrow and "Ads Agent = the chat" is too narrow. The owner of this fact is this file.
+- RULE: the conversational experience is a closed beta. "Enrolled beta advertisers can describe what they want to accomplish in their own words, get tailored recommendations and actions, and review and approve changes before they're made." "This experience is in closed beta and continues to actively evolve. Availability may vary by eligibility, capability, marketplace, language, and capacity." Access is through an account executive. Countries listed: CA, MX, US, BR, CO, CL, BE, DE, ES, FR, IT, NL, PL, SE, TR, UK, EG, SA, AE, AU, IN, JP, SG, NG, ZA. [official] (*Ads Agent: An AI marketing assistant from Amazon Ads*, product page, https://advertising.amazon.com/solutions/products/ads-agent, read 2026-11-09); unBoxed 2026 calls it "now available in expanded closed beta" [official] (*The 11 biggest takeaways from unBoxed 2026*, above).
+- RULE: what it does for sponsored ads. "For sponsored ads, advertisers can apply keyword, bid, and budget recommendations with one click, then review and approve changes." It also plans media (audience sizing, reach forecasts, budget recommendations), turns a media plan into campaigns, makes "bulk changes across campaigns at once within DVA+ advanced settings", reports with "trend diagnosis, benchmarking, and visual analysis", writes AMC SQL, and answers billing and ad-policy questions. "Advertisers get to review a summary of proposed changes and approve them before they're applied". [official] (product page, above)
+- Amazon's numbers, vendor-reported beta comparisons with no holdout described *(for human readers only; section M)*: "During beta, advertisers saw 40% faster first sales with auto-optimized Sponsored Products campaigns" (footnote 1: "Amazon internal data, July–Sep 2026, US. Based on beta estimates; results are not predictive of future results."); applying targeting recommendations in natural language gave "more than 25% additional unique customers on their Display, Video, and Audio campaigns" (footnote 2: "Amazon internal data, Jan–Aug 2026, WW. Based on beta estimates; results are not predictive of future results."), and the news adds "while lowering costs by more than 10% for cost-per-impression on their campaigns on average". unBoxed 2026 adds "nearly 1,000 metrics" queryable in natural language with peer benchmarks. [official] (product page, footnotes re-read in the gap fill; *Introducing Amazon Ads Agent*; unBoxed 2026 takeaways)
+  - [seat] The "auto-optimized Sponsored Products campaigns" of footnote 1 match, by name and by the 40% figure, the auto-optimization that Amazon Ads' small-business page and the Accelerate 2026 recap describe (section L). The pages do not say whether the three 40% figures come from one beta.
+- RULE: the 2025 starting point. At unBoxed 2025 Amazon unified "the Ads Console and Amazon DSP into a single media buying tool" called Campaign Manager and introduced Ads Agent: "our Ads Agent tool actions at the direction of advertisers" (sic) and "is accessible in a conversational, chat window on certain pages"; a user can "Upload a custom media plan and let Ads Agent create your campaign structure and ad groups", "optimize your campaigns at scale using only natural language", take audience and keyword recommendations to "review, refine, and apply", and build AMC SQL. [official] (*unBoxed 2025: Simplifying advertising through innovation*, expert advice, 2025-11-11, Kelly MacLean, https://advertising.amazon.com/library/news/unboxed-2025-simplifying-advertising-ai, read 2026-11-09)
+- [seat] No Ads API page found on 2026-11-09 exposes Ads Agent's conversation, its proposals or its approvals. The agent's sponsored-ads recommendations are, by Amazon's description, the same keyword, bid and budget recommendations section D reads; what is applied through it lands on the campaigns as ordinary changes.
+
+### H. The Amazon Ads MCP server
+
+- RULE: what it is. The server exposes two primitives, Tools and Skills. Standard tools are "Deterministic actions"; Expert tools have "Inference behind the tool". Amazon "does not own or operate the LLMs" a client uses. An example prompt on the page is "Increase my campaign budget to $500 on `[campaign_id]`". [official] (*Overview* (Amazon Ads MCP Server), https://advertising.amazon.com/API/docs/en-us/mcp/mcp-overview, markdown at https://d3a0d0y2hgofx6.cloudfront.net/en-us/mcp/mcp-overview.md, read 2026-11-09)
+- RULE: access. Endpoints `https://advertising-ai.amazon.com/mcp`, `-eu` and `-fe`; the same Login with Amazon client and tokens as the Ads API, or OAuth 2.1 sign-in (2026-07-06: Kiro CLI, Claude Code, Claude Desktop, ChatGPT). Account context is Dynamic, the default ("the LLM asks you to provide the account identifiers needed to invoke each selected tool"), or Fixed (`Amazon-Ads-AI-Account-Selection-Mode: FIXED` with at least one of `Amazon-Advertising-API-Scope`, `Amazon-Ads-AccountID` or `Amazon-Ads-Manager-AccountID`). In the Kiro CLI configuration a client can hide tools with `disabledTools`, the page's example being `["campaign_management-delete_campaign", "campaign_management-delete_ad"]`. [official] (*Connecting to the Amazon Ads MCP Server*, https://advertising.amazon.com/API/docs/en-us/mcp/get-started; release notes; read 2026-11-09). [seat] This file's seat did not call any MCP endpoint.
+- RULE: the Lite form, served at `/mcp/lite` from 2026-09-29, has three tools, `amazon_ads_mcp-search_tools`, `amazon_ads_mcp-describe_tools` and `amazon_ads_mcp-invoke_tool`, used as search, describe, invoke; it is "functionally equivalent to the Regular version", and a search returns the whole domain namespace. [official] (*Connecting to the Amazon Ads MCP Server (Lite)*, https://advertising.amazon.com/API/docs/en-us/mcp/get-started-lite; release notes 2026-09-29; read 2026-11-09)
+- RULE: the campaign management tools. Query tools for campaigns, ad groups, ads, targets, ad associations and portfolios; create, update and delete for campaigns, ad groups, ads, targets and ad associations; `update_campaign_state`, `update_campaign_budget` (budgets with `recurrenceTimePeriod` `DAILY`, `LIFETIME` or `MONTHLY`), `update_target_bid`; `create_portfolio`, `create_singleshot_portfolio`, `update_portfolio`; `add_country_campaign` (Sponsored Products manual, first 10 ASINs); `check_product_eligibility`; `create_bid_adjustment_rule` (DSP); `dsp_create_conversion_tracking_products`; and `create_singleshot_sp_campaign`. [official] (MCP tool specification `campaign_management.json`, https://d3a0d0y2hgofx6.cloudfront.net/en-us/mcp/campaign_management.json, read 2026-11-09)
+- RULE: the single-shot campaign tool. It creates "ONE global campaign with country-specific budgets"; targeting is automatic by default, and `autoCreateTargets` false makes it manual; target types are `KEYWORD`, `PRODUCT`, `PRODUCT_CATEGORY`, `THEME` and `LOCATION`; the default name is "SP|Auto|Campaign [timestamp]"; `oneshotCampaigns` takes one item. 28 of the 32 tool inputs (all but the four portfolio tools) carry an optional `skill` object (`skillName`, `version`) with the instruction "Do not auto-generate or infer values for this object." [official] (same specification)
+- RULE: other tool domains: reporting (`create_campaign_report`, `create_report`, `retrieve_report`, `delete_report`, `create_inventory_report`, `create_product_report`), account management, ads accounts, eligibility, billing, audience discovery (`get_audience`, `list_audiences`, `list_audience_categories`, added 2026-10-30), AMC (query execution and workflows, 2026-05-19), Amazon Live, inventory management, manager accounts, promotions, terms token, test accounts, advertiser product group eligibility, and users, user invitations, user permissions and user roles (the specification files are `users.json`, `user_invitations.json`, `user_permissions.json`, `user_roles.json`; there is no `user_management.json`). The AMC SQL Generator (`amc_sql_generator`) is an Expert tool in closed beta. [official] (MCP tool specifications; *AMC SQL generator*, https://advertising.amazon.com/API/docs/en-us/mcp/amc-sql-generator; release notes; read 2026-11-09)
+- RULE: the open beta note (2026-02-18) promised campaign management, reporting and account management tools, a "single-shot campaign creation tool", and workflows to "automate keyword harvesting from automatic to manual campaigns" and for "country expansion". [official] (release notes, read 2026-11-09). [seat] The current `campaign_management.json` has the country tool but no harvesting tool; the harvesting workflow is not visible in the public specification on 2026-11-09.
+
+### I. The MCP skills library and the migration skills
+
+- RULE: skills. "Workflow Skills" chain several tools for a task; "Capability Skills" teach a client one domain; clients such as Claude Code trigger them automatically. They are downloaded with `GET https://advertising-api.amazon.com/skills?name=all&download=true` with the same headers, and can be unzipped into `.claude/skills/`. "Skills are supported in AI applications that recognize Anthropic skill definitions." [official] (*Overview* (MCP skills), https://advertising.amazon.com/API/docs/en-us/mcp/skills/skills-overview; *Download Skills from the Amazon Ads MCP Server*, https://advertising.amazon.com/API/docs/en-us/mcp/skills/get-started; read 2026-11-09). [seat] The download is an Ads API host and was not called by this seat; the skill contents are known only from their library pages.
+- RULE: the library's published skills on 2026-11-09: reporting (`ads-reporting`), advertiser accounts, advertiser product group eligibility, AMC workflows, account context identifiers, manager accounts, product eligibility, program eligibility, terms token, user invitations and user permissions (the last three added 2026-07-21), and `dsp-campaign-analysis` as an example workflow. A further workflow skill, *Create DSP P+/B+ Campaign*, creates and launches a DSP Performance+ or Brand+ campaign from one request: it gathers inputs, creates the campaign, sets ASIN conversion tracking, queries eligible tactics, creates ad groups and activates the campaign. None is about Sponsored Products bids, budgets, harvesting or optimization. [official] (MCP skills library pages under https://advertising.amazon.com/API/docs/en-us/mcp/skills/skills-library/, including *Create DSP P+/B+ Campaign Skill* (`create-dsp-pb-campaign`), read 2026-11-09)
+- RULE: the AI migration skill `ads-api-migration` (2026-11-02) helps a developer move code to the unified v1 API; Amazon's public repository `amzn/ads-advanced-tools-docs` ships `unified-campaign-management-migration-skills/skills.json` listing `unified-api-migration-guide`, `unified-sp-migration`, `unified-sb-migration`, `amazon-ads-sb-collections`, `unified-api-cli-testing`, `amazon-ads-spglobal`, `unified-dsp-cm-migration`, `amazon-ads-dsp-creative-management` and `update-migration-skills`; the same repository holds generative-AI samples, Marketing Stream CloudFormation templates and Postman collections. [official] (*AI Migration Skills*, https://advertising.amazon.com/API/docs/en-us/mcp/ai-migration-skills; https://github.com/amzn/ads-advanced-tools-docs; read 2026-11-09)
+- RULE: Amazon's engineering description of agents in advertising: a router agent over domain agents, agent-to-agent (A2A) plus MCP, built on Bedrock AgentCore. [official] (*AI Agents in Advertising*, Ads API knowledge hub blog, Chintan Sanghavi, 2025-12-27, https://d3a0d0y2hgofx6.cloudfront.net/en-us/knowledge-hub/blogs/genai/gen-ai-amazon-ads.md, read 2026-11-09). Ten hands-on MCP workshops sit under the knowledge hub (listed, not read).
+
+### J. Amazon Marketing Stream: the data automation runs on
+
+The data surface is `method/reporting.md`'s; the lines here are why automation depends on it.
+
+- RULE: Stream v2 (GA 2026-09-30) delivers to the advertiser's own AWS: Parquet files in an S3 bucket with an SQS notification. `ads-performance-v1` carries about 400 metrics at hourly grain, delivered "as often as every 15 minutes"; `ads-converted-product-v1` arrives every 4 to 6 hours. Records are totals, not deltas: keep the highest `streamBatch.version`; delivery is at least once with no ordering; history backfills within about a day. Limits include 5 destinations per application, 25 requests a second and files of about 256 MB *(for human readers only)*. Subscriptions are managed with `POST /adsApi/v1/create|query|retrieve|delete/streamSubscriptions`. [official] (Marketing Stream v2 pages `guides/amazon-marketing-stream/v2/overview`, `v2/datasets/overview`, `v2/manage-subscriptions/limits` and `v2/data-delivery` under https://advertising.amazon.com/API/docs/en-us/; release notes 2026-09-30; v1 merged OpenAPI path list; read 2026-11-09)
+- RULE: v1 sent hourly delta JSON through Firehose and SQS and also carried messaging and recommendation datasets; it stops on 2027-06-30 and its messaging and recommendation data are not carried into v2 (section D). [official] (*Deprecations*, above)
+- RULE: the two v1 recommendation datasets that end with Stream v1. `sponsored-ads-campaign-diagnostics-recommendations` (beta) pushed recommendation types such as "Campaign with over 50% budget utilization, which will be impacted by an upcoming special event ... Apply the recommended budget", "Low-performing Sponsored Brands campaign with at least one impression in the last 14 days and ROAS less than or equal to 3. Apply the recommended bid", and Sponsored Display creation prompts; `sp-budget-recommendations` pushed `suggested_budget`, `reason_for_suggestion`, missed sales, clicks and impressions bounds over a 7-day window and `percent_time_in_budget` ("-1 means there is not enough information"). Both pages carry the June 30, 2027 shutoff warning. [official] (*Sponsored ads campaign recommendations (beta)*, https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/datasets/sponsored-ads-campaign-diagnostics-recommendations; *Sponsored Products budget recommendations and missed opportunities*, https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/datasets/budget-recs-missed-opportunities; read 2026-11-09). The thresholds in those type descriptions are Amazon's triggers for its own suggestions, *(for human readers only)*.
+- RULE: Reporting v1 conversions are on traffic date. [official] (release notes 2026-09-30, read 2026-11-09)
+- RULE, with Amazon's pages in conflict: Reporting v1's concurrency scope. Per advertiser: the GA note ("The Reporting API v1 supports one concurrent report per advertiser, which allows advertisers to run approximately 100 reports per day"), the *Amazon Ads reporting overview* ("one concurrent report per advertiser. Only one report can be running at a time for a given advertiser account") and the Stream v2 *Overview* ("subject to usage limits (one concurrent report per advertiser, approximately 100 reports per day)"). Per client id: the *Reporting API v1 overview* ("one concurrent report per client ID, not per advertiser account") and *Limits and best practices* ("Reporting API v1 applies its rate and concurrency limits per client ID ... Limits are not applied per advertiser account", concurrency "as low as 1"). [official] (release notes; `guides/reporting/overview`; `guides/amazon-marketing-stream/v2/overview`; `guides/reporting/ads-v1/overview`; `guides/reporting/ads-v1/limits`; all re-opened as cloudfront markdown 2026-11-09 in the gap fill) The figures are *(for human readers only)*; the scope is `method/reporting.md`'s (P-reporting-15), and no page reconciles it.
+- [seat] Every hourly tool below (Teikametrics' hourly bidding, Perpetua's conversion optimization "using Stream data", Quartile's hourly placements, dayparting in Helium 10 and Scale Insights) depends on Stream's hourly grain. Stream needs an AWS destination the seller owns; the open door can create a subscription but cannot read an S3 bucket.
+
+### K. What Amazon says about letting it decide
+
+- GUIDANCE: Amazon's framing of control: "Get more done without giving up control ... Review recommendations and proposed changes before they're applied, approve what works, and refine the rest." [official] (Ads Agent product page)
+- GUIDANCE: on testing: "limit your changes while testing strategies, so that you can attribute the difference in performance to a specific change." [official] (dynamic bidding guide)
+- GUIDANCE: on rule targets: set the ROAS guardrail from Amazon's recommendation, because "if you set your ROAS guardrails too aggressively, you may see a decrease in the number of conversions"; for Sponsored Brands cost rules, adjust the target or budget if it is missed "over a 21-day period (excluding special days)". [official] (rule-based bidding overview; SB optimization rules guide)
+- RULE: Amazon's own estimates are labelled as estimates: budget recommendation ranges are "not guaranteed"; theme bid ranges are "impact metrics"; "Amazon Ads does not guarantee the ROAS guardrail". [official] (SP 3.0 OpenAPI; rule-based bidding overview)
+
+### L. Seller Assistant (formerly Project Amelia) and auto-optimization: Amazon's second agent
+
+Added in the gap fill. Seller Assistant is the Seller Central agent; Amazon Ads Agent (section G) is the Amazon Ads platform. Both now reach a seller's sponsored ads.
+
+- RULE, history. 2024-09-19: "Today, we are introducing in beta a new generative AI-powered personal assistant for sellers, codenamed Project Amelia", "available from any page within Seller Central", built on Amazon Bedrock; "As Amelia evolves, it will ... increasingly gain the ability to not only converse with sellers, but anticipate their needs, take actions, and resolve issues on their behalf." [official] (*Amazon launches a powerful new generative AI-based selling assistant codenamed Project Amelia*, aboutamazon, Mary Beth Westmoreland, https://www.aboutamazon.com/news/innovation-at-amazon/amazon-project-amelia, read 2026-11-09). 2025-09-17 (Accelerate 2025): "Last year, we introduced Seller Assistant ... Today, I'm excited to announce the next evolution of Seller Assistant, enhanced with agentic AI capabilities ... to reason, plan, and help take action with a seller's permission"; it runs on Amazon Bedrock "leveraging Amazon Nova and Anthropic Claude"; on ads, it extends "agentic AI capabilities to advertising through Creative Studio" and "will also be able to ... recommend optimized advertising and marketing strategies"; "Seller Assistant is currently available to all sellers in the U.S. store ... at no additional cost." [official] (*Amazon introduces agentic AI across the seller experience, transforming how sellers manage their businesses* (HTML title "Amazon unveils new agentic AI-powered Seller Assistant to help independent sellers grow"), aboutamazon, https://www.aboutamazon.com/news/innovation-at-amazon/seller-assistant-agentic-ai, read 2026-11-09; *Amazon Accelerate 2025: A new era of tools and technology for independent sellers*, https://sellingpartners.aboutamazon.com/amazon-accelerate-2025-a-new-era-of-tools-and-technology-for-independent-sellers, read 2026-11-09: "Seller Assistant can set goals, develop strategies, and with permission, take action on behalf of sellers")
+  - [seat] Amazon's own pages date the first Seller Assistant three ways: 2023 ("We first launched Seller Assistant in 2023", 2026-09-23 article), 2024 (the 2025 article's "Last year", matching Project Amelia's 2024-09-19 beta) and 2025 ("first debuted at Accelerate a year ago", Accelerate 2026 recap). The dated history is not load-bearing; what it does now is.
+- RULE, Accelerate 2026 on ads (2026-09-29 recap). The assistant "checks whether each product's listing and inventory are actually ready to advertise, flags what needs fixing, and then drafts Sponsored Products and Sponsored Brands campaigns the seller can approve in a keystroke or open in the ad console to fine-tune." Then: "A new auto-optimization capability keeps working while the seller is doing something else: adding search terms that are performing, pausing those that aren't, and holding every bid under a maximum the seller sets. In early results, Amazon said, new campaigns using it saw a first sale 40 percent faster." Muthu Muthukrishnan, vice president of Sponsored Products and Brands, is quoted: "You can always see exactly what changed and why, and make any adjustments." [official] (*At Accelerate 2026, Amazon shares new AI tools as a force multiplier for independent sellers*, sellingpartners.aboutamazon.com, `datePublished` 2026-09-29, read 2026-11-09)
+- RULE, the same capability on Amazon Ads' own page (2026-09-22). "Auto-optimization continuously finds and acts on opportunities ... Powered by AI, auto-optimization expands product targeting to help advertisers engage with audiences who are considering similar or complementary products and categories, adds relevant keywords and pauses underperforming ones, and identifies seasonal opportunities—all within the advertiser's existing campaign settings." "Enable auto-optimization with a single click in your campaign settings. Once turned on, Amazon Ads continuously discovers high-performing keywords and product targets on your behalf. Advertisers can see how each action performs and edit or reverse actions at any time". "In beta testing, campaigns that adopted keyword auto-optimization saw a first sale 40% faster and a 13% increase in sales" *(for human readers only; section M)*. "Auto-optimization will be available to advertisers in the United States in January 2027." The same page announces an "AI-powered campaign intelligence" dashboard on the campaign overview page that "might flag that a top-performing keyword is losing impression share and recommend a bid adjustment", where "advertisers can apply, edit, or skip each action", "available to all sponsored ads advertisers in the United States in October 2026". [official] (*Amazon Ads introduces new sponsored ads tools to help small businesses achieve their goals*, expert advice, 2026-09-22, https://advertising.amazon.com/library/expert-advice/sponsored-ads-small-business-tools, read 2026-11-09)
+- RULE, how much Seller Assistant does without asking, in Amazon's words, which do not agree. The recap: "Sellers choose whether it surfaces insights, drafts actions, or acts on its own", and Joy Chang "described a model in which the seller defines the job and sets the guardrails, with nothing happening without approval and every action logged." The workflows article: "For any workflow, prebuilt or custom, sellers choose whether it just surfaces recommendations or takes actions on their behalf. When it takes actions, sellers stay in control, reviewing and approving them before they're carried out"; workflows "run continuously ... even when the seller isn't logged in to Seller Central", with "a complete audit trail"; examples are pricing, restocking, listings and ratings, not ads. Acceptance: "sellers accepting recommendations over 90% of the time" *(for human readers only)*. [official] (Accelerate 2026 recap; *Amazon gives sellers an even smarter Seller Assistant and a new plugin for Amazon Quick and Anthropic's Claude*, aboutamazon, 2026-09-23, https://www.aboutamazon.com/news/innovation-at-amazon/seller-assistant-plugin-amazon-quick-claude, read 2026-11-09)
+  - [seat] Read together: drafted campaigns wait for the seller's keystroke; auto-optimization, once switched on, adds keywords and product targets and pauses keywords without a per-change approval, inside the campaign's settings and under the seller's maximum bid, and each change can be reversed afterwards. Amazon states no threshold for "performing" or "underperforming", and no page read names the maximum-bid field or says whether the maximum is per target or per campaign.
+- RULE, the external plugin. "Through a new plugin, sellers can connect their Seller Central data to outside AI tools they already use, starting with Anthropic's Claude and Amazon's own Quick in the U.S." (recap); "Their AI doesn't just read from their Amazon business; it can act on it, the same way Seller Assistant does inside Seller Central", with "human approval on actions, and complete audit trails"; "now available in beta for sellers in Amazon's U.S. stores". [official] (2026-09-23 article; Accelerate 2026 recap)
+- RULE, no API surface found. On 2026-11-09 the Ads API release notes contain no entry for auto-optimization or Seller Assistant, and the v1 merged model's campaign schema carries only `autoCreationSettings` (`autoCreateTargets`, "Gives Amazon permission to automatically create targets associated with the campaign based on the products being advertised"; `autoManageCampaign`) and no auto-optimization or maximum-bid field. [official] (release notes; v1 merged OpenAPI, schemas `Campaign`, `AutoCreationSettings`; re-read 2026-11-09 in the gap fill)
+  - [seat] Whether the console's auto-optimization switch is `autoCreateTargets`, `autoManageCampaign` or a field not yet in the model is not stated on any page read. The targets it adds would be ordinary targets on the campaign, read through `POST /adsApi/v1/query/targets`; the change history API does not say who made a change (section D, P-22).
+- [seat] How it differs from Ads Agent, by Amazon's pages: Seller Assistant is a Seller Central tool for sellers, free, available to US sellers and rolled out to "over 90% of Amazon's selling partners worldwide"; it drafts SP and SB campaigns from listing and inventory readiness, which Amazon Ads Agent's pages do not describe; Ads Agent's conversational experience is a closed beta reached through an account executive, covers sponsored ads, DVA and Full-Funnel, and applies "keyword, bid, and budget recommendations with one click". Both name the same SP auto-optimization and the same 40% figure. Neither has a public API to its conversation.
+
+### M. How Amazon's own automation figures were measured
+
+Added in the gap fill. Every lift figure in sections D to L is Amazon's report on its own product. None is a figure a skill uses as an expected effect for one seller.
+
+- RULE, what the footnotes say. Ads Agent 40% faster first sales and 25% more unique customers: "Based on beta estimates; results are not predictive of future results" (product page). Full-Funnel 67% higher Long-Term ROAS: "against the blended average of other campaign performance across beta advertisers ... Early beta advertiser data; individual results may vary" (*Grow total sales with Full-Funnel Campaigns*). Keyword auto-optimization 40% and 13%: "In beta testing, campaigns that adopted keyword auto-optimization" (small-business page). Creative tools in Ads Agent: "up to 71% more product coverage and up to 15% sales growth versus their results before adopting the suite" (unBoxed 2026). Campaign presets 29%: "compared to similar \* advertisers", "Amazon internal data, WW, September 2022 – August 2023" (Partner Opportunities). SB headline optimization 1.6%: "Based on internal testing, headline text optimization can drive up to 1.6% `CTR` lift vs. non-optimized campaigns" (release notes, 2025-04-17). Ads data manager: "advertisers using first-party data with Amazon Ads see a 30% increase in ROAS compared to those who don't" (small-business page). Prompts: shoppers who click "convert to a sale 48% more often ... than those who don't" (same page). [official] (pages as named, read 2026-11-09)
+  - [seat] The comparison groups are adopters against non-adopters, adopters against their own past, or a blended average of other campaigns. None of these pages describes a randomised holdout. `method/change-measurement.md` P-42 records the same for DCO and Full-Funnel; Improvado's Ivo Petrovic reads the Full-Funnel footnote as "not a holdout test on your account" [opinion] (*Amazon unBoxed 2026: What Changed for Retail Media Metrics*, https://improvado.io/blog/amazon-unboxed-2026-retail-media-measurement, read 2026-11-09).
+- RULE, a selection rule built into one automation. Rule-based bidding is open only to a campaign that has been "running for at least 10 days prior to setting a rule and ha[s] a minimum of 10 conversions in the last 30 days". [official] (*Sponsored Products rule-based bidding overview*, section B)
+  - [seat] So every campaign switched to rule-based bidding through the API was already converting, and a comparison of those campaigns with campaigns that were not eligible mixes the rule's effect with the gate's. A campaign switched on after a strong month is also exposed to regression to the mean (next item), which makes a before/after look worse than the rule's effect; a campaign switched on after a weak month is exposed the other way.
+- [research] Regression to the mean: "unusually large or small measurements tend to be followed by measurements that are closer to the mean", and the effect "becomes more noticeable with increasing measurement error and when follow-up measurements are only examined on a sub-sample selected using a baseline value". (Barnett, van der Pols and Dobson, *Regression to the mean: what it is and how to deal with it*, International Journal of Epidemiology 34, 2005, pp. 215–220, doi 10.1093/ije/dyh299; abstract read on PubMed 2026-11-09)
+- [research] Interference in an auction. In Baidu's AiAds study the 670 treated advertisers were chosen by Baidu from 13,350 that had set up conversion tracking ("we choose 670 advertisers to use the AiAds system as the experimental group, and the others can be treated as the controlled group"), and the authors write: "Since the advertising platform is an auction market, the performance of AiAds will gradually abate as the increase of more advertisers who choose the AiAds". That is the authors' expectation, not a measured decline. (Yang et al., *AiAds*, KDD 2019, https://arxiv.org/abs/1907.12118, full text read 2026-11-09)
+- [research] Test-control interference measured. Blake and Coey found test-control interference in an eBay email-marketing experiment, and "naive estimates of the TATE ignoring test-control interference exaggerate the treatment's effectiveness by about a factor of 2"; Fradkin's simulation of marketplace search experiments overstates true effects "by over 100%". (Blake and Coey, *Why marketplace experimentation is harder than it seems: the role of test-control interference*, ACM EC 2014, pp. 567–582, doi 10.1145/2600057.2602837, Crossref record read 2026-11-09; their result as summarised by Holtz and Aral, *Limiting Bias from Test-Control Interference in Online Marketplace Experiments*, 2020, https://arxiv.org/abs/2004.12162, full text read 2026-11-09. The ACM page answered a browser challenge and the paper itself was not read.)
+  - [seat] Transfer to Amazon: when a beta's treated campaigns bid in the same auctions as untreated ones, part of the measured gain can be taken from the comparison group, so the gain shrinks when every advertiser in the auction has the same automation. The figure an Amazon beta reports is closest to "a few early campaigns, against everyone else", not "this seller's campaign once the feature is the default".
+
+## What practitioners add
+
+Everything in this section is [opinion]: what a vendor or practitioner says its product does, from its own public pages unless the line says a search snippet. A vendor's own result figures are marketing, kept *(for human readers only)*. Pricing is context, read from the page named, and changes often.
+
+### What the tools automate, in one view
+
+[seat] The seat's grouping of the vendor pages below. "Amazon native" names the Amazon mechanism that does the same job, from the section above.
+
+| Job | Amazon native | Third-party tools that state they do it |
+|---|---|---|
+| Set a bid per target from a target ACoS or ROAS | rule-based bidding (campaign level, ROAS); `PRIORITIZE_KPI_TARGET` (v1) | Helium 10 Ads, Perpetua, Teikametrics, Scale Insights, m19, Ad Badger, Quartile, Pacvue (snippet), Skai, BidX, Adbrew, Intentwise |
+| Move bids within the day (dayparting) | schedule bid rules (raise only) | Helium 10 Ads, Scale Insights, Teikametrics (hourly), Quartile, Pacvue (hourly up and down, Stream), Skai (Stream), Perpetua (hourly with Stream) |
+| Change placement adjustments | placement recommendations (suggest only) | Scale Insights (placement algorithm), Quartile (hourly), Adbrew |
+| Harvest search terms into manual targets | target promotion groups; auto-optimization (US, January 2027) | Helium 10 Ads, Scale Insights (import), Perpetua (not in conversion mode), Adbrew, m19, Pacvue (snippet), Skai |
+| Add negatives from wasted spend | negative brand recommendations only | Helium 10 Ads, Scale Insights (negative, negative word, blacklist), Ad Badger, Adbrew, Skai |
+| Raise or lower budgets on rules | budget rules (raise only), unspent sharing | Helium 10 Ads, Pacvue, Skai (Budget Navigator), Scale Insights (budget), Intentwise |
+| Pause and revive targets | none in the API rules (they cannot pause); auto-optimization pauses keywords (announced, US January 2027) | Scale Insights (revive), Helium 10 Ads, Pacvue |
+| Build campaigns from a template | single-shot MCP tool, Ads Agent, Full-Funnel, Seller Assistant drafts (SP, SB) | Helium 10 Ads ("60 seconds"), Perpetua, m19, Teikametrics |
+| Defend organic rank with bids | none | Helium 10 Ads, Perpetua (rank-based rules), Pacvue (snippet) |
+| An agent connection (MCP) | Amazon Ads MCP server | Helium 10, Intentwise, Ad Badger, Adbrew, Pacvue, Skai (Agent Connect), several open-source servers |
+
+### Helium 10 Ads (formerly Adtomic)
+
+- [opinion] Helium 10's ads tool is now sold as "Helium 10 Ads" at the former Adtomic address (the pages read do not state the rename or the ownership; Helium 10's link to Pacvue is not sourced here). Its page promises: "Launch campaigns in 60 seconds and let AI manage the rest"; "Bid Rules: Raise or lower bids automatically to hit your ACoS target"; "Dayparting automatically shifts your budget toward the hours and days that produce real revenue" and "Schedule bids to activate or pause at specific hours"; rules over "ACoS thresholds, bid formulas, negative targeting criteria"; "Smart Suggestions surface high-converting search terms, flag underperforming targets, and recommend bid changes"; "Keyword Harvesting: Add winning search terms into exact match campaigns automatically"; "Negative Targeting: Block underperforming search terms"; "Budget Rules: Auto-adjust daily budgets based on clicks, spend, or day of week"; and "Keyword Rank Rules: Defend your rank by auto-adjusting bids when rank changes in search results". (Helium 10, *Helium 10 Ads*, https://www.helium10.com/tools/advertising/adtomic/, read 2026-11-09)
+- [opinion] Its pricing page lists, for the Diamond tier ("$100K – $10M in yearly sales"), "Helium 10 MCP: Included free (up to 1,000 calls)" and "Rules-based ad automation and dayparting"; "Helium 10 Ads is available in Diamond plans" and "Diamond plan customers incur a 2% management fee on PPC spend managed through Helium 10 Ads" *(for human readers only)*. (Helium 10, pricing page, https://www.helium10.com/pricing/, read 2026-11-09)
+- [opinion] Helium 10's MCP blog lists tools such as `get_keywords_by_asin` and example workflows including a "Nightly worst-campaign flag" (the skill "pulls prior-day campaign data via Ads MCP → ranks by ACoS → isolates the bottom 3 → Claude diagnoses the root cause ... → sends one recommended fix per campaign"). Its free "Claude skills" page offers a Keyword Opportunity Finder skill, a Listing Analyzer skill, a Brand Analytics Keywords skill and a Search Query Performance skill. (Helium 10, MCP blog, https://www.helium10.com/blog/helium-10-mcp/ and *Helium 10 MCP: the only marketplace intelligence MCP with data that Amazon can't give you*, https://www.helium10.com/blog/helium-10-mcp-the-only-marketplace-intelligence-mcp-with-data-that-amazon-cant-give-you/; *Claude skills*, https://www.helium10.com/claude-skills/; read 2026-11-09)
+- [opinion] From search snippets of Helium 10's knowledge base, whose pages answered 403 to the seat: a default target ACoS of 30% for new campaigns, and accepted bid suggestions applied within 24 hours. Not verified on the page itself.
+
+### Perpetua
+
+- [opinion] Method, in Perpetua's help centre: the ad engine aims the aggregate ACoS at the target and "prioritizes sales over ACoS". "Going dark mitigation" lowers the effective target ACoS in the background when a campaign would otherwise stop serving. (Perpetua, *What is the ad engine doing?*, https://help.perpetua.io/en/articles/5897118-what-is-the-ad-engine-doing, read 2026-11-09)
+- [opinion] Two optimisation modes for Sponsored Products: ACoS optimisation ("Bids are calculated daily, or when a change is made by the user"); conversion optimisation ("Optimization is done hourly using Stream data") against an "ACoS Ceiling", "a guardrail for the optimizer" that "will mitigate further bid increases on targets that reach the ceiling"; conversion mode "supports 'Keyword Targeting' campaign creation only" and "Harvesting is not supported". Perpetua recommends setting the ceiling "slightly above your actual target (we recommend 1.3* Target ACoS)". (Perpetua, *Perpetua's optimisation for Sponsored Products (open beta)*, https://help.perpetua.io/en/articles/10545202-perpetua-s-optimisation-for-sponsored-products-open-beta, read 2026-11-09)
+- [opinion] From search snippets of Perpetua help only: the target should be lowered by no more than a relative 10% at a time. Not found on the pages read.
+- [opinion] Perpetua's homepage shows rules on organic rank, for example "Organically ranked as #1, decrease spend". (Perpetua, https://perpetua.io/, read 2026-11-09)
+- Pricing: "$695" for "Up to 10k in monthly ad spend"; "$695 /mth + % of ad spend" over 10k *(for human readers only)*. (Perpetua, *Perpetua Pricing & Plans*, https://perpetua.io/pricing/, read 2026-11-09 by the verifier)
+
+### Teikametrics (Flywheel)
+
+- [opinion] Method: "The optimal bidding strategy in a second-price auction is to bid the true value of the ad placement to the advertiser." A new target gets a starter bid, then a discovery stage ("On average, the discovery stage can take 15 days"), then a value-based bid: "(payoff per order) x (conversion rate) x (Goal) X (ACoS Limit) = Value Based Bid". "In auto campaigns, bids are calculated at the ad group level and in manual campaigns at the target level." (Teikametrics help, *Decoding predictive AI bidding*, https://help.teikametrics.com/en/articles/5644036-decoding-predictive-ai-bidding-a-guide-to-optimizing-your-advertising-strategy; Flywheel FAQ, https://intercom.help/flywheel-20/en/articles/5644025; read 2026-11-09)
+- [opinion] Bids are evaluated "multiple times throughout the day"; hourly bidding uses Amazon's data feed and changes bids "up to 6 times more frequently". (Teikametrics help, *Hourly bidding*, https://intercom.help/flywheel-20/en/articles/7030461, read 2026-11-09)
+- [opinion] "ARI Assistant is the agentic AI that works alongside your team ... ranking the highest-impact moves"; "Every recommendation is prioritized by impact and applied in one click". (Teikametrics, *Flywheel*, https://teikametrics.com/flywheel, read 2026-11-09)
+- Pricing (search snippets): an Essentials plan at about $149 to $179 a month up to $10K ad spend, then about 3% of spend *(for human readers only)*.
+
+### Scale Insights
+
+- [opinion] Named algorithms: Bidding, Dayparting, Import (harvesting), Negative, Negative Word, Blacklist, Whitelist, Revive, Default Bidding (legacy auto campaigns), Status (pause or enable campaigns and ad groups), Daily Budget and Placement; "All algorithms run once a day, except Dayparting Rules, which run hourly". (Scale Insights, https://scaleinsights.com/; *Overview of Each Automation Algorithm*, https://docs.scaleinsights.com/docs/overview-of-each-algorithm.md; index https://docs.scaleinsights.com/llms.txt; read 2026-11-09)
+- [opinion] The bid formula: "`Optimum Bid = (Keyword Revenue / Clicks) * Target ACOS`", revenue per click times the target; the page names two further formulas, one from the associated ASIN's selling price, target ACoS and conversion rate, and one for keywords with clicks and no sales. The tool normalises for placement modifiers and the dynamic up/down strategy, because "Amazon can increase your bid based on your campaign's placement modifiers and bidding strategy". The analysis window is set by "Days of Recent Data to Ignore" and "Days to Analyze" (the page's example: 2 and 7), available "up to 180 days" for most algorithms and "up to 365 days" for the Bidding Rule. "Hourly bid adjustments are cleared when the base bid is updated", and "Bidding rules always calculate using the current base bid, not the hourly-adjusted bid." (Scale Insights documentation: *How to calculate optimum bid*, https://docs.scaleinsights.com/docs/how-to-calculate-optimum-bid.md; *Bidding algorithm: normalize bids based on placement and dynamic up/down*, https://docs.scaleinsights.com/docs/bidding-algorithm-normalize-bids-based-on-placement-and-dynamic-updown.md; *How do algorithms analyze data*, https://docs.scaleinsights.com/docs/how-do-algorithms-analyze-data.md; *Considerations for hourly bidding*, https://docs.scaleinsights.com/docs/considerations-for-hourly-bidding.md; read 2026-11-09). [seat] The documentation's worked example mixes a $300 and a $500 revenue figure, so the example is not reliable; the formula is.
+- [opinion] The negative-word rule's example filter is at least 25 clicks and 0 orders *(for human readers only)*. (Scale Insights, *What is the negative word rule*, https://docs.scaleinsights.com/docs/what-is-the-negative-word-rule.md, read 2026-11-09)
+- Pricing: "1% of monthly ad spend" for unlimited automated ASINs *(for human readers only)*. (Scale Insights homepage, https://scaleinsights.com/, read 2026-11-09)
+
+### m19
+
+- [opinion] "m19 automates bid optimization, keyword discovery and campaign structure"; Autopilot and Professional "run on the same machine learning algorithms". (m19, https://www.m19.com/, read 2026-11-09)
+- Pricing, from m19's own comparison post: "Autopilot $59/mo · Professional $479 + 3% of sponsored ad spend"; the same post gives Perpetua as "From ~$695/mo; % of spend above $10K" *(for human readers only)*. (m19, *m19 vs Perpetua*, https://www.m19.com/blog/m19-vs-perpetua, read 2026-11-09 on the page itself by the verifier). The pricing page answered "in progress" to the research seat.
+
+### Quartile
+
+- [opinion] "Dynamic bidding based on real-time data", "Automated placement adjustments by the hour", and "dedicated account support that includes a team of experts working proactively to strategize, unlock opportunities, and optimize". (Quartile, https://www.quartile.com/, read 2026-11-09)
+
+### Intentwise
+
+- [opinion] "AI Gateway", an MCP server for Claude and Copilot with a semantic layer over the seller's advertising and retail data. (Intentwise, *AI Gateway*, https://www.intentwise.com/ai-gateway/, read 2026-11-09)
+- [opinion] An earlier Intentwise post (2021-07-30, Aditi Sharma, *Amazon Budget Rules – A Boon for the Advertisers?*) explains Amazon's budget rules and how to use them around events. (Intentwise, *Amazon budget rules*, https://www.intentwise.com/blog/amazon-budget-rules, read 2026-11-09). Its help article on harvesting with automated rules (https://help.intentwise.com/en/articles/5170239-harvest-keywords-and-targets-with-automated-rules) answered 404.
+
+### Ad Badger
+
+- [opinion] "Our bid optimization tool uses a proprietary bid algorithm to analyze your conversion data, respond daily with micro bid adjustments, and hit your target ACOS"; "Our negative keyword tool scans every single keyword in your ad campaign over the past six months"; and "Connect Claude & ChatGPT to your Amazon Ads". The blog's pre-built negative rules blacklist, over six months, a keyword "with over $35 in ad spend without conversions" or "over 34 clicks without a single conversion" *(for human readers only)*. (Ad Badger, https://www.adbadger.com/; *The Easiest Way to Automate Negative Keywords for Amazon Ad Campaigns*, https://www.adbadger.com/blog/amazon-ppc-education/automate-negative-keywords/; read 2026-11-09. The *Bid optimization* page, https://www.adbadger.com/bid-optimization/, answered 404 to the verifier; the bid claim stands on the homepage.)
+
+### BidX and Adbrew
+
+- [opinion] BidX sells AI agents that propose actions and "execute the ones you approve". (BidX, https://www.bidx.io/, read 2026-11-09). A BidX blog post on Amazon's dynamic bidding and placement bids (https://bidx.io/blog/amazons-dynamic-bidding-strategies-placement-bids) was listed in a search and not read.
+- [opinion] Adbrew: "Adbrew MCP is now available for all customers"; "Optimize bids, budgets, and placements with hourly performance insights". (Adbrew, https://www.adbrew.io/, read 2026-11-09)
+
+### Pacvue
+
+Pacvue's own site (pacvue.com, including *Pacvue for Amazon*, its blog and its customer stories) answered a Cloudflare challenge (403) to curl and to the fetch tool on 2026-11-09, and the Internet Archive was not reachable from this seat. What follows is Pacvue's own words on public copies: its wire press releases and its case studies as republished on casestudies.com. Pacvue's help centre is not public.
+
+- [opinion] Pacvue Agent (2026-04-14): "Brand and agency teams can interact with Pacvue Agent through a conversational, natural-language interface or directly in Slack. Through agentic workflows, recommendations can be converted into approval-based execution with defined guardrails"; teams can "Convert insights into campaign and budget updates through agentic workflows with built-in guardrails, approvals, and auditability" and "Translate business questions into AMC-ready queries". Sunava Dutta, chief product officer: "Pacvue Agent is not a stand-alone chatbot layered onto reporting". (Pacvue press release, *Pacvue Launches Pacvue Agent*, as carried by Destination CRM, https://www.destinationcrm.com/Articles/CRM-News/CRM-Across-the-Wire/Pacvue-Launches-Pacvue-Agent-174367.aspx, read 2026-11-09). Luis Rijo (PPC Land) reports that at launch it "initially covers Amazon Ads exclusively", lists "governed execution" (a recommendation becomes "a bid change, a budget adjustment, a targeting modification" through "guardrails, approval steps, and an audit trail"), and records Pacvue's figures "workflows up to 200x faster, 80x quicker time to insight, and 54% performance gains", which he later describes as "from early adopters without disclosed methodology" *(for human readers only)*. (PPC Land, *Pacvue Agent promises 200x faster commerce media workflows*, 2026-04-15, https://ppc.land/pacvue-agent-promises-200x-faster-commerce-media-workflows/; *Pacvue Prism launches today to kill the channel budget debate*, 2026-06-22, https://ppc.land/pacvue-prism-launches-today-to-kill-the-channel-budget-debate/; read 2026-11-09)
+- [opinion] Pacvue Prism (2026-06-22): "a commerce-native platform" around an "Agentic Commerce Grid", "a brand's unified view of media performance: every channel connected to underlying commerce signals", connecting to "data lakes & warehouses, CDPs, CRMs, and custom AI agents via MCP". Rijo reports that Prism's "proprietary halo measurement" is "built on regression modeling combined with custom attribution and pixel tracking" and that Pacvue launched an MCP server on 2026-05-14 connecting its data to ChatGPT, Copilot, Gemini and Claude. (Pacvue press release, GlobeNewswire, as carried by WBOC, https://www.wboc.com/online_features/press_releases/pacvue-launches-pacvue-prism-powering-the-industrys-first-agentic-commerce-grid/article_e81a6b7d-95a8-5dcb-8bbd-97da57251b7b.html; PPC Land, above; read 2026-11-09)
+- [opinion] Dynamic dayparting, in Pacvue's case studies: "Pacvue implemented Dynamic Dayparting, powered by Amazon Marketing Stream data, to automate hourly bid adjustments based on historical 14-day performance patterns. By increasing bids during high-CVR hours and reducing them during low-CVR periods"; for OLLY, "automated hourly bid adjustments based on conversion rates", where the problem was "campaigns going offline during peak shopping hours". The result figures (31% revenue, 26% ROAS; 3% lower CPC, 510 basis points of conversion rate) are Pacvue's and *(for human readers only)*. (*A Leading Beverage Manufacturer Achieves 31% Revenue Growth with Pacvue's Dynamic Dayparting* and *OLLY reduces CPC by 3% with Pacvue*, Pacvue case studies on casestudies.com, https://www.casestudies.com/company/pacvue/case-study/a-leading-beverage-manufacturer-achieves-31-revenue-growth-with-pacvues-dynamic-dayparting and https://www.casestudies.com/company/pacvue/case-study/olly-reduces-cpc-by-3-with-pacvue, read 2026-11-09)
+  - [seat] Unlike Amazon's schedule bid rules (raise only), Pacvue's dayparting lowers bids in weak hours as well; it does so by hourly writes, which is the scheduler section "What an agent with the open door already does" names.
+- [opinion] Search snippets only, not verified on a Pacvue page: the *Pacvue for Amazon* page's "Manage Sponsored Search and DSP with rules-based automation and intelligent bid optimization"; a Pacvue blog rule "When my product's organic search rank =1; pause bids"; hourly share-of-voice analysis from AMC feeding dayparting; a "dynamic model" refreshed daily from "the last 14 days (excluding the latest 2 days)"; a rule convention that a threshold "must hold for 7 days (excluding the last 3)"; Pacvue Commerce as the retail-analytics module (sales, inventory, digital shelf). Pricing is custom.
+- [seat] What Pacvue automates by default cannot be read from its public pages on 2026-11-09: the sources above describe options a user configures (rules, dayparting) and an agent whose changes wait for approval.
+
+### Skai
+
+- [opinion] Retail media features, on Skai's own page: "Powerful machine learning continuously adjusts bids, budgets, and targeting to maximize ROI in real time"; "Skai Portfolios & Budget Navigator: Implement daily bid and budget optimizations with powerful algorithms to make data-driven and effective decisions based on your goals"; "Search Term Analysis: Automated negative keyword analysis and implementation"; "AI Dayparting: Translate Amazon's hourly Stream data into impactful optimizations with automated workflows to efficiently manage intraday bidding"; "Advanced Automated Actions: Change settings across your campaigns, ad groups, keywords, and ads according to your parameters ... specify multiple conditions all at once"; "Automated Keyword Harvesting ... based on your desired conditions"; "Competitive Insights: Extract channel share of voice insights on your brand and competing brands". (Skai, *Retail Media Solutions*, https://skai.io/retail-media/, read 2026-11-09)
+- [opinion] Incrementality: the product page now titled "Marketing Measurement Solution in the Era of Privacy" (address `/impact-navigator/`, and its customer quotes still say "Impact Navigator") offers "customizable incrementality testing", "Aggregate data only. No tracking, third-party cookies, or personally identifiable information", results "in weeks, not months"; it names no Amazon-specific method. (Skai, https://skai.io/impact-navigator/, read 2026-11-09). Skai's blog calls holdout testing "a gold standard", "going 100% dark (i.e., pausing spending) on a test group while leaving the rest of your spending (i.e., control group) as-is", and its author reports "it's common to see 5-15% of ad sales attributed to non-branded campaigns from branded search queries" in Amazon accounts audited. (Skai, *Retail Media Incrementality: Skai's Innovative Approach*, 2023-07-27, https://skai.io/blog/retail-media-incrementality-skais-innovative-approach/, read 2026-11-09). Skai's Amazon incrementality route is AMC (*Unlocking Incrementality on Amazon with Amazon Marketing Cloud via Skai*, Ana Alicia Santaella, 2023-06-08, https://skai.io/blog/unlocking-incrementality-on-amazon-with-amazon-marketing-cloud-via-skai/, read 2026-11-09).
+  - [seat] A geo-split test like the Privalia one Skai describes needs ads that can be switched off by region. No page this file read describes regional targeting for Sponsored Products inside one marketplace (the v1 and MCP target types include `LOCATION`, whose sponsored-ads use no page read explains), so on Amazon the splits this file can source are by product, by campaign or by time, which is `method/change-measurement.md`'s subject.
+- [opinion] Agents: Skai Agent (Celeste AI) inside the platform, and Agent Connect (2026-09-22), which lets outside agents "create and modify campaigns, adjust bids and budgets, change campaign status and edit large numbers of objects across multiple accounts. Every proposed change remains subject to human review and approval, with a clear audit trail." (Skai press release, *Skai Launches Agent Connect, Opening Its Omnichannel Marketing Infrastructure to External AI Agents*, https://skai.io/press-releases/skai-launches-agent-connect-opening-its-omnichannel-marketing-infrastructure-to-external-ai-agents/, read 2026-11-09; Skai homepage, https://www.skai.io/, read 2026-11-09)
+- [seat] As with Pacvue, Skai's public pages describe configurable automation (budget navigator, automated actions, dayparting, harvesting) and do not state which run by default; its help centre is not public.
+
+### Sellozo, Jungle Scout
+
+- [opinion] Sellozo: its pages are a JavaScript shell; a third-party snippet gives a $250 a month platform fee *(for human readers only)*.
+- [opinion] Jungle Scout's features page (Jungle Scout Catalyst) shows product and keyword research and, under "Review Automation and AI Assist", "Automate PPC campaigns to save time and resources, while gaining valuable insights and data-driven recommendations"; it states no bid method. (Jungle Scout, https://www.junglescout.com/features/, read 2026-11-09)
+
+### Open-source and community MCP servers and skills
+
+- [opinion] Public GitHub repositories wrap the Ads API as MCP servers or Claude skills: KuudoAI/amazon_ads_mcp, MarketplaceAdPros/amazon-ads-mcp-server, Xnurta-MCP, nospicyplease/amazon-ppc-advanced-skills, Deltologic/datadoe-mcp, agentcentral-to. (GitHub, read 2026-11-09). [seat] They are the same shape as `cx-amazon-ads`: a credential, a thin door and instructions.
+
+### What practitioners agree on, and where they split
+
+- [opinion] Shared method across Helium 10, Scale Insights, Teikametrics, Ad Badger and Perpetua: the bid is derived from a value per click (revenue per click, or conversion rate times average order value) times a target ACoS, recomputed on a schedule; targets with too little data get a starting bid and a discovery period; search terms that convert are moved to exact match and terms that spend without orders become negatives.
+- [opinion] They split on frequency (daily for Perpetua's ACoS mode and Ad Badger; several times a day for Teikametrics; hourly for dayparting tools), on the lookback (Scale Insights' user-set days to ignore and to analyse, its example 2 and 7; Pacvue's 7 skipping 3, per snippet only), and on whether harvesting runs inside the bidder (Perpetua's conversion mode does not).
+- [opinion] Several vendors have moved from "AI does it" to "AI proposes, you approve" (BidX, Ads Agent's approval step), and from a dashboard to an MCP connection (Helium 10, Intentwise, Ad Badger, Adbrew).
+- [seat] Every third-party tool writes through the same Ads API endpoints section D and the door use. What a tool adds is the schedule (it runs without being asked), a store of history longer than Amazon's report retention, and its bid formula; none of them has an Amazon write the door lacks.
+
+### What an agent with the open door already does that these tools do
+
+[seat] The door is `cx raw amazon-ads --origin <origin> --method <m> --path <p> [--header ...] [--body ...]` against `advertising-api.amazon.com`, `-eu` or `-fe`, with the profile scope header or `Amazon-Ads-AccountId` for v1 and cross-product calls (`cx-amazon-ads` skill and its `references/door.md`). The door adds no wait to a 429, reads a lost reply to a POST list as "sent, outcome unknown", and a report's download url is fetched outside it. Against that, job by job:
+
+| Job a tool sells | Reachable through the door | What the door does not give |
+|---|---|---|
+| Read every campaign, ad group, target, bid, budget, strategy and automation flag | yes: v3 list calls and `POST /adsApi/v1/query/*` | nothing missing |
+| Set or move a bid, a placement adjustment, a strategy, a budget | yes: v3 update calls, `POST /adsApi/v1/update/*` | a scheduler: the change happens when cx is asked or run |
+| Create and read Amazon's own rules (rule-based bidding, schedule and v2 optimization rules, budget rules) | yes: `/sp/rules/*`, `/sp|sb|sd/budgetRules` | SP optimization rules and budget rules only raise; rule-based bidding moves bids both ways toward Amazon's reading of the ROAS goal; a seller-chosen decrease is a direct write |
+| Dayparting | yes, by writing Amazon's schedule rules (Amazon then runs them) | down-moves inside the day need hourly writes, which needs a scheduler |
+| Read Amazon's recommendations and apply one | yes: `/sp/campaign/recommendations`, bid, budget, keyword recommendations, `/adsApi/v1/query/recommendations`; apply is the write the recommendation names | Ads Agent's chat proposals have no API |
+| Harvest search terms into manual targets | yes: Reporting v1 or v3 search term reports, then target writes; or Amazon's `/sp/targetPromotionGroups` | the report download step is outside the door |
+| Add negatives | yes: negative keyword and target writes | same report step |
+| Build a campaign set from a template | yes: v1 create calls or the v3 sequence | none |
+| Opt out of auto-assembled videos | yes: `POST /adsApi/v1/update/campaigns` with `optOuts` | none |
+| Hourly data | partly: a Stream v2 subscription can be created | the S3 bucket and SQS queue are the seller's AWS, not reachable through the door |
+| Keep a long history of bids and results | no store of its own | report retention is Amazon's; a change log is Amazon's change history API (`POST /history`), which covers SP and SB only and "doesn't return information about who made the change" |
+| Organic rank inputs for rank-based rules | not on the Ads API | rank comes from other sources (SP-API, listing tools) |
+| Amazon's MCP skills | the `/skills` path sits on the Ads API origin | not called by this seat; whether the door's caller is allowed to fetch it is open |
+
+The difference between cx and a bid tool is therefore the clock and the store, not the reach: a tool runs every hour without being asked and keeps its own history; cx acts when a seller's conversation or a scheduled run invokes it, and reads history from Amazon each time. Amazon's own rule engines (schedule bid rules, budget rules, rule-based bidding, target promotion groups) are the part of "automation without a clock" the door can hand to Amazon to run. [seat]
+
+## Published research
+
+Only what the seat opened is cited; for most items that is the abstract or the bibliographic record, and the line says so.
+
+- [research] Platform-provided bid recommendations, as evidence. In Zillow's switch from negotiated contracts to auctions, many bidders did not use the platform's recommended bid; inferring values from bid histories, "for half of the agents not following the recommendation, the increased effort of experimenting with alternate bids results in increased regret", and the share not following "slowly declines as markets mature, but it remains large"; the authors attribute it to "the lack of trust in the platform-provided tool". (Jalaly, Nekipelov and Tardos, *Learning and Trust in Auction Markets*, 2017, https://arxiv.org/abs/1703.10672, abstract read 2026-11-09)
+- [research] Autobidders under budget and ROI constraints. A gradient-based autobidder that uses only bandit feedback can satisfy budget and return-on-investment constraints with vanishing regret, and when such autobidders compete, "the resulting expected liquid welfare over all rounds is at least half of the expected optimal liquid welfare", whether or not bidding converges. (Lucier, Pattathil, Slivkins and Zhang, *Autobidders with Budget and ROI Constraints: Efficiency, Regret, and Pacing Dynamics*, 2023, https://arxiv.org/abs/2301.13306, abstract read 2026-11-09). [seat] This is the theory behind a target-ROAS or target-ACoS strategy like `PRIORITIZE_KPI_TARGET` or rule-based bidding: the constraint is met by pacing a multiplier, not by a per-bid rule.
+- [research] Amazon's own work on learning to bid. Amazon scientists frame bidding as a bandit or reinforcement-learning problem, because "an advertiser does not observe counterfactual outcomes of bid amounts that were not submitted, and successful advertisers will adapt their own strategies based on bids placed by competitors", and release AuctionGym, a simulator for offline evaluation of bidding strategies. (Jeunen, Murphy and Allison, *Learning to bid with AuctionGym*, AdKDD 2022, https://www.amazon.science/publications/learning-to-bid-with-auctiongym, abstract read 2026-11-09; code https://github.com/amazon-science/auction-gym, last pushed 2026-02-11). [seat] The counterfactual point is why neither a third-party bidder nor cx can read "what would this target have done at another bid" from a report; it can only change the bid and watch, which is `method/change-measurement.md`'s problem.
+- [research] A platform's automated bidding at scale. Baidu's AiAds system replaced manual keyword bidding with automated bidding ("the advertisers adopt the target CPA bidding language ... their keyword-level manual bidding is invalid"), broad match with intelligent targeting, and creative generation. In an online A/B test against manual optimisation, AiAds raised conversions 56.91% and the platform's revenue 43.51% (table 2; the text says "more than 47% improvement in the revenue"); in a year-long grouping experiment (last week of 2018 against last week of 2017), the 670 treated advertisers, chosen by Baidu from 13,350 with conversion tracking, grew revenue 27.22% and budget 30.20%, against 10.85% and 8.83% for the rest. The authors expect that "the performance of AiAds will gradually abate as the increase of more advertisers who choose the AiAds", because the platform "is an auction market". (Yang et al., *AiAds: Automated and Intelligent Advertising System for Sponsored Search*, KDD 2019, https://arxiv.org/abs/1907.12118; full text read 2026-11-09 in the gap fill, section 7 and tables 2–3) *(figures for human readers only)*. [seat] "Revenue" in table 2 is the sponsored search system's revenue, that is Baidu's; the grouping experiment is not randomised; section M carries what this means for Amazon's own beta figures.
+- [research] Test-control interference and regression to the mean, the two reasons a beta figure overstates or misstates one seller's expected effect: Blake and Coey (ACM EC 2014; via Holtz and Aral 2020) and Barnett, van der Pols and Dobson (2005). They are quoted in section M with their records.
+- [research] Platform-side objectives. Large e-commerce ad platforms run the sell side, the buy side and the exchange themselves, so "the goal of the auto-bidding algorithms is not only to maximize the advertisers' conversions, but also the total revenue of the platform"; the benchmark tests bidders under homogeneous, heterogeneous and promotion-event competition such as Black Friday budget surges. (Yang et al., *PlatformBid: An Auto-Bidding Benchmark from a Unified Advertising Platform's Perspective*, 2026, https://arxiv.org/abs/2607.27265, abstract read 2026-11-09). [seat] Amazon is such a platform; its "spend double" warning on up and down and its unspent-budget sharing are consistent with a bidder that serves both sides, which is a reason a seller's own view of value per click matters.
+- [research] Human reliance on algorithms. People "erroneously avoid algorithms after seeing them err" (Dietvorst, Simmons and Massey, *Algorithm aversion*, Journal of Experimental Psychology: General, 2015, doi 10.1037/xge0000033), while in other tasks "people prefer algorithmic to human judgment" (Logg, Minson and Moore, *Algorithm appreciation*, Organizational Behavior and Human Decision Processes, 2019, doi 10.1016/j.obhdp.2018.12.005). (Crossref records, titles read 2026-11-09). [seat] The two findings bracket why sellers both hand campaigns to tools and turn them off after one bad week.
+- [research] Not found on 2026-11-09: a peer-reviewed or working paper that measures the effect of Amazon's own dynamic bidding, rule-based bidding, or a third-party Amazon bid tool on seller outcomes. Practitioner tests on Google Ads disagree on manual versus automated bidding (Inflow and Marketing Syrup posts, seen only as search summaries), and are not cited as evidence here. An Osmos survey (search summary only) reports that more than four in five advertisers use automated bidding, an adoption figure, not an effect.
+- [research] Paid search effects in general (Blake, Nosko and Tadelis, *Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field Experiment*, Econometrica, 2015, doi 10.3982/ECTA12423) are `areas/ads-organic-and-total.md`'s subject; listed here only because automation that maximises attributed sales inherits that paper's caution about incrementality.
+
+## Stable enough for a skill
+
+Each item is a fact the seat expects to hold for months, with the pages it stands on. Figures that change are not here; they are in the next section.
+
+1. Amazon runs several automations on an advertiser's campaigns, each with its own switch: dynamic bidding strategies, rule-based bidding, schedule and optimization rules, budget rules, portfolio unspent-budget sharing, automatic targeting, keyword groups, Sponsored Brands theme targeting and optimization rules, the v1 flags `autoCreateTargets`, `autoManageCampaign`, automatic budget allocation and automatic global scaling, auto-assembled videos, headline optimization and prompt ads. (SP 3.0 OpenAPI; v1 merged OpenAPI; *Sponsored Products auto targeting*; *Sponsored Product keyword groups targeting*; *Sponsored Brands theme-based targeting*; *Managing Sponsored Brands optimization rules*; *Auto-assembled videos*; *Prompt Ad Extension reports*; release notes)
+2. Under "up and down" Amazon's current help, library guide and API models say it moves a bid by up to 100% in either direction for all placements, while Amazon's author's guide still says up to 100% for top of the first page and up to 50% elsewhere (the owner is `areas/bids-and-placements.md`, P-8); under "down only" it never raises it; under "fixed" it does not move it. (*Bidding strategies for Sponsored Products*; *Guide to dynamic bidding - up and down*; SP 3.0 OpenAPI `BiddingStrategy`; v1 merged OpenAPI `BidStrategy`; *An author's guide to Sponsored Products*)
+3. Sponsored Products optimization rules (schedule and performance) and budget rules only add a percentage to a bid or a budget; rule-based bidding and Sponsored Brands optimization rules move bids down as well as up on Amazon's reading of performance; no Amazon rule pauses a target. (SP 3.0 OpenAPI `OptimizationRulesAPISwagger*` schemas; *Schedule Bid Rules*; *Sponsored ads budget rules overview*; *Bidding rules for Sponsored Products*; *Sponsored Products rule-based bidding overview*; *Managing Sponsored Brands optimization rules*)
+4. Rule-based bidding is a campaign-level ROAS guardrail that Amazon does not guarantee, can raise a bid up to 5 times the adjusted bid, cannot run alongside schedule bid rules, and is switched off by Amazon when ROAS collapses against its earlier level, restoring the previous strategy and bids. (*Sponsored Products rule-based bidding overview*; *Getting started with the Sponsored Products rule-based bidding API*; *Bidding rules for Sponsored Products*)
+5. Budget rules only raise a daily budget, are evaluated at the start of the day and on change, and add up when several apply. (*Sponsored ads budget rules overview*; *Budget rule evaluation*; *Performance-based rules*; *Schedule-based rules*)
+6. With unspent-budget sharing enabled, a campaign in a portfolio can spend up to twice its daily budget from its portfolio's unspent budget, within the monthly portfolio budget. (release notes, 2025-07-19)
+7. Amazon's recommendations are estimates refreshed about daily, and applying one is the ordinary write to the object it names. (SP 3.0 OpenAPI, Consolidated Recommendations and budget recommendation descriptions; *Recommendations API overview*; *How to call the API*; *Partner Opportunities - Campaign Optimization*)
+8. The cross-product Recommendations API returns types the reader may not know, and a reader is told to tolerate them. (*Recommendations API overview*; *Recommendation types*)
+9. Target promotion groups are Amazon's own harvesting from an auto ad group into manual keyword and product ad groups, with recommendations based on historical performance. (*Target Promotion*; SP 3.0 OpenAPI `TargetPromotionGroups`; *Partner Opportunities - Campaign Optimization*)
+10. Auto-assembled videos, assembled by Amazon from existing uploaded video clips, are on for every US Sponsored Products campaign unless the advertiser opts out, and the opt-out does not pause videos already delivering. (*Auto-assembled videos*)
+11. Amazon Ads Agent is the name, from 2026-09-29, of Amazon's unified ads platform (sponsored ads, Display, Video, and Audio, Full-Funnel Campaigns); its conversational experience, a closed beta, proposes changes and applies them after the advertiser approves, and has no public API; the help centre's "formerly known as Amazon DSP" is the same platform name written into DSP-specific passages. (*Introducing Amazon Ads Agent*; *Ads Agent: An AI marketing assistant from Amazon Ads*; *Understand delivery alerts and recommendations*; *Understand Advertiser Accounts*)
+12. The Amazon Ads MCP server is a separate host from the Ads API with its own tool names; its Lite form exposes search, describe and invoke over the same tools; its skills are downloaded from a path on the Ads API origin. (*Overview* (Amazon Ads MCP Server); *Connecting to the Amazon Ads MCP Server*; *Connecting to the Amazon Ads MCP Server (Lite)*; *Overview* (MCP skills); *Download Skills from the Amazon Ads MCP Server*; MCP tool specifications)
+13. Marketing Stream v2 delivers hourly performance data to the advertiser's own S3 bucket and SQS queue; v1, with its recommendation and messaging datasets, stops on 2027-06-30. (Marketing Stream v2 pages; *Deprecations*; v1 dataset pages)
+14. Third-party bid tools derive a bid from value per click times a target ACoS on a schedule and write it through the Ads API. (Scale Insights *How to calculate optimum bid*; Teikametrics *Decoding predictive AI bidding*; Perpetua *What is the ad engine doing?*; Helium 10 *Helium 10 Ads*; Ad Badger *Bid optimization*)
+15. A bid's counterfactual result at another bid is not observable in a report. (*Learning to bid with AuctionGym*)
+16. Existing sponsored ads campaigns are automatically eligible to serve in Alexa for Shopping and to be optimized as prompts, with no documented campaign switch; individual prompts can be paused in the console. (*What agentic shopping means for advertising*; *Prompt Ad Extension reports*)
+17. Amazon's change history API returns SP and SB changes but not who made them, so it cannot by itself separate a seller's edit from a tool's or Amazon's. (Change history OpenAPI, `POST /history`)
+18. Seller Central's Seller Assistant (Project Amelia in its 2024 beta) is a second Amazon agent that reaches sponsored ads: it drafts Sponsored Products and Sponsored Brands campaigns for the seller to approve, and Amazon describes an SP auto-optimization that, once switched on, adds keywords and product targets and pauses keywords without a per-change approval, under a maximum bid the seller sets. (*At Accelerate 2026, Amazon shares new AI tools ...*; *Amazon Ads introduces new sponsored ads tools to help small businesses achieve their goals*; *Amazon launches ... Project Amelia*)
+19. Amazon's published lift figures for its own automation compare adopters with non-adopters, with their own past or with a blended average, from beta or internal data, and none of the pages describes a randomised holdout. (Ads Agent product page footnotes; *Grow total sales with Full-Funnel Campaigns*; small-business tools page; unBoxed 2026; release notes 2025-04-17; *Partner Opportunities - Campaign Optimization*)
+20. Rule-based bidding admits only campaigns with at least 10 days running and 10 conversions in 30 days (API guide), so the campaigns on it were already converting when it was switched on. (*Sponsored Products rule-based bidding overview*)
+21. Reporting v1's concurrency is stated per advertiser on three Amazon pages and per client id on two. (release notes 2026-09-30; *Amazon Ads reporting overview*; Stream v2 *Overview*; *Reporting API v1 overview*; *Limits and best practices*)
+
+## What cx must read at run time
+
+The figures and states a skill must fetch, not remember:
+
+- Each campaign's bidding strategy, placement adjustments, `goalSettings` and automation flags (`autoCreateTargets`, `autoManageCampaign`, `budgetAllocation`, `marketplaceBudgetAllocation`, global scaling), and `optOuts`: `POST /adsApi/v1/query/campaigns` or the v3 list calls.
+- Rules attached to each campaign and their status: `POST /sp/rules/optimization/search` (with a campaign filter; there is no `GET` on `/sp/campaigns/{campaignId}/optimizationRules`), `POST /sp/rules/campaignOptimization/state`, `GET /{sp|sb|sd}/campaigns/{campaignId}/budgetRules`, `POST /sb/rules/optimization/list`, and `POST /sp/rules/campaignOptimization/eligibility` with its suggested guardrail.
+- Portfolio `budgetControls` and unspent-budget sharing.
+- Recommendations as of today: `GET` or `POST /sp/campaign/recommendations` (the two answer different media types and shapes), budget recommendations, theme-based bid recommendations, target promotion recommendations, `POST /adsApi/v1/query/recommendations` in discovery mode with its live type list (and `GET /adsApi/v1/recommendations`, the `ListRecommendation` path in the v1 model, which the guide writes `/adsApi/v1/list/recommendations`).
+- Budget usage: `POST /sp|sb|sd/campaigns/budget/usage`.
+- Event ids and dates for schedule rules: `POST /sp/v1/events` and budget-rule recommended events.
+- Marketplace availability lists (intraday budget rules, keyword groups, auto-assembled videos, Video Generator, Ads Agent countries), minimum daily budgets for rule-based bidding, rule count limits (20 schedule bid rules, 100 rules per campaign, 250 budget rules, 25 ids per association, 300 rules per create), Stream limits, Reporting v1's concurrency (whose scope, per advertiser or per client id, Amazon's pages state both ways; a `429` or queued answer is the live reading): all from the live pages or the live error, since each is dated *(for human readers only)* above.
+- What changed on a campaign and when: the change history API (`POST /history`, `Changehistory_prod_3p.json`, see `method/ads-api-surface.md`; `THEME` events need `Accept: application/vnd.historyresponse.v1.1+json`). It covers Sponsored Products and Sponsored Brands, not Sponsored Display, and "doesn't return information about who made the change"; who changed it (seller, tool, Amazon) is read from the automation switches above, not from this API.
+- `optOuts` and the `videoType` of each SP ad extension (`POST /adsApi/v1/query/adExtensions`), to see which auto-assembled videos are delivering.
+- The MCP specification and skills library, if cx is to compare with or borrow Amazon's skills: their content changes release to release.
+- Whether auto-optimization or a Seller Assistant draft is behind a campaign: no Ads API field found on 2026-11-09. What can be read is its result: targets on the campaign (`POST /adsApi/v1/query/targets`, or the v3 keyword and target lists), their states, the campaign's `autoCreationSettings`, and the change history (`POST /history`), which does not name who made a change. Whether the switch is visible in the API after the January 2027 US release is read from the reference and the release notes then.
+- The current wording of Amazon's own figures and their footnotes, if one is quoted to a seller: from the page itself, since footnotes change with each beta (Ads Agent product page, *Grow total sales with Full-Funnel Campaigns*, the small-business tools page).
+- Deprecation dates (v3 reporting and Stream v1 on 2027-06-30; SB benchmark APIs 2026-12-17; SB v2 impression share returning NULL from 2026-11-10): from *Deprecations* at run time.
+
+## Pages read
+
+Ads API guide pages are listed at their public `advertising.amazon.com/API/docs/en-us/` address; the seat read each as the public markdown copy the documentation site serves from `https://d3a0d0y2hgofx6.cloudfront.net/en-us/<path>.md`, navigated with that site's `llms.txt`, `toc2.json` and `searchIndex.json`. All read 2026-11-09.
+
+| Page | URL | Type | Date read | What it gave |
+|---|---|---|---|---|
+| Introducing Amazon Ads Agent (news) | https://advertising.amazon.com/library/news/amazon-ads-agent | official | 2026-11-09 | Ads Agent as the unified platform; Full-Funnel, DVA+; one-click apply; beta figures |
+| Ads Agent: An AI marketing assistant from Amazon Ads | https://advertising.amazon.com/solutions/products/ads-agent | official | 2026-11-09 | closed beta, approval step, countries, 40% faster first sales |
+| The 11 biggest takeaways from unBoxed 2026 | https://advertising.amazon.com/library/news/unboxed-2026-news-announcements | official | 2026-11-09 | expanded beta, Full-Funnel figures, creative figures, MCP Lite, 1,000 metrics |
+| unBoxed 2025: Simplifying advertising through innovation | https://advertising.amazon.com/library/news/unboxed-2025-simplifying-advertising-ai | official | 2026-11-09 | Campaign Manager, first Ads Agent, Creative Agent streaming TV |
+| Transform campaign and asset development with Creative Agent (what's new) | https://advertising.amazon.com/resources/whats-new/unboxed-2025-creative-agent | official | 2026-11-09 | no-cost agentic creative, US, who can use it |
+| Amazon Ads launches new agentic AI tool that creates professional-quality ads | https://advertising.amazon.com/library/news/amazon-ads-agentic-ai-creative-tool | official | 2026-11-09 | the creative agent's steps and outputs |
+| Video Generator is now available for Sponsored Brands video campaigns in nine marketplaces | https://advertising.amazon.com/resources/whats-new/video-generator-for-sponsored-brands-video-campaigns | official | 2026-11-09 | six 15-second videos, countries |
+| AI creative solutions (Creative Solutions - Optimize your creative with AI-powered solutions) | https://advertising.amazon.com/generative-ai-ad-solutions | official | 2026-11-09 | the creative tool family and availability |
+| What agentic shopping means for advertising (news, added by the verifier) | https://advertising.amazon.com/library/news/agentic-shopping-advertising | official | 2026-11-09 | Rufus and Alexa+ merged into Alexa for Shopping; automatic eligibility; prompts; Alexa+ conversational ads |
+| Bidding rules for Sponsored Products (help GDLQ5D2BNFCAU3TW, added by the verifier) | https://advertising.amazon.com/help/GDLQ5D2BNFCAU3TW | official | 2026-11-09 | 5× cap, up and down, exclusion with schedule rules, console eligibility |
+| Bidding strategies for Sponsored Products (help GCU2BUWJH2W3A8Z7, added by the verifier) | https://advertising.amazon.com/help/GCU2BUWJH2W3A8Z7 | official | 2026-11-09 | 100% for all placements |
+| Sponsored Brands 3.0 OpenAPI (added by the verifier) | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/SponsoredBrands_prod_3p.json | official | 2026-11-09 | `/sb/recommendations/optimization`, SB rule paths |
+| Change history OpenAPI (added by the verifier) | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/Changehistory_prod_3p.json | official | 2026-11-09 | `POST /history`; SP and SB only; no who-changed field |
+| Guide to dynamic bidding - up and down with Sponsored Products | https://advertising.amazon.com/library/guides/dynamic-bidding-sponsored-products | official | 2026-11-09 | 100% all placements, "spend double", two-week review, limit changes |
+| Ads API release notes | https://advertising.amazon.com/API/docs/en-us/release-notes/index | official | 2026-11-09 | dated entries 2023-2026: MCP, Recommendations API, Stream v2, unspent sharing, DCO, audiences, Rufus, Alexa+, Review Requests |
+| Deprecations | https://advertising.amazon.com/API/docs/en-us/release-notes/deprecations | official | 2026-11-09 | 2027-06-30 shutoffs; recommendation data moves to the Recommendations API |
+| Overview (Amazon Ads MCP Server) | https://advertising.amazon.com/API/docs/en-us/mcp/mcp-overview | official | 2026-11-09 | tools and skills; standard vs expert tools |
+| Connecting to the Amazon Ads MCP Server | https://advertising.amazon.com/API/docs/en-us/mcp/get-started | official | 2026-11-09 | endpoints, headers, account modes, `disabledTools` |
+| Connecting to the Amazon Ads MCP Server (Lite) | https://advertising.amazon.com/API/docs/en-us/mcp/get-started-lite | official | 2026-11-09 | three meta-tools |
+| Overview (MCP skills) | https://advertising.amazon.com/API/docs/en-us/mcp/skills/skills-overview | official | 2026-11-09 | workflow vs capability skills |
+| Download Skills from the Amazon Ads MCP Server | https://advertising.amazon.com/API/docs/en-us/mcp/skills/get-started | official | 2026-11-09 | `/skills` download path |
+| MCP skills library: reporting, advertiser accounts, advertiser product group eligibility, AMC workflows, account context identifiers, manager accounts, product eligibility, program eligibility, terms token, user invitations, user permissions, create DSP P+/B+ campaign | https://advertising.amazon.com/API/docs/en-us/mcp/skills/skills-library/ | official | 2026-11-09 | what Amazon's published skills cover |
+| AI Migration Skills | https://advertising.amazon.com/API/docs/en-us/mcp/ai-migration-skills | official | 2026-11-09 | `ads-api-migration` |
+| AMC SQL generator | https://advertising.amazon.com/API/docs/en-us/mcp/amc-sql-generator | official | 2026-11-09 | expert tool, closed beta |
+| MCP tool specification: campaign management | https://d3a0d0y2hgofx6.cloudfront.net/en-us/mcp/campaign_management.json | official | 2026-11-09 | tool list, single-shot campaign, `skill` object |
+| MCP tool specifications: reporting, account management, ads accounts, eligibility, billing, audience discovery, AMC, Amazon Live, inventory, manager accounts, promotions, terms token, test accounts, users | https://d3a0d0y2hgofx6.cloudfront.net/en-us/mcp/<domain>.json | official | 2026-11-09 | the other tool domains |
+| amzn/ads-advanced-tools-docs | https://github.com/amzn/ads-advanced-tools-docs | official | 2026-11-09 | migration skills list, samples, Stream templates, Postman |
+| AI Agents in Advertising (knowledge hub blog) | https://d3a0d0y2hgofx6.cloudfront.net/en-us/knowledge-hub/blogs/genai/gen-ai-amazon-ads.md | official | 2026-11-09 | router and domain agents, A2A, AgentCore |
+| Recommendations API overview | https://advertising.amazon.com/API/docs/en-us/guides/recommendations/recommendations-api/overview | official | 2026-11-09 | object shape, states, tolerate unknown types |
+| How to call the Recommendations API | https://advertising.amazon.com/API/docs/en-us/guides/recommendations/recommendations-api/how-to | official | 2026-11-09 | GET and POST, headers, discovery mode, example responses |
+| Recommendation types | https://advertising.amazon.com/API/docs/en-us/guides/recommendations/recommendations-api/recommendation-types | official | 2026-11-09 | the four types, `INVALID_RECOMMENDATION_TYPE` |
+| Partner Opportunities - Campaign Optimization | https://advertising.amazon.com/API/docs/en-us/guides/recommendations/partner-opportunities/catalog/campaign-optimization | official | 2026-11-09 | AT-to-MT over 64 days, presets, refresh daily |
+| Sponsored Products 3.0 OpenAPI | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/SponsoredProducts_prod_3p.json | official | 2026-11-09 | strategies, recommendations, rules v1 and v2, target promotion groups, budget endpoints |
+| Amazon Ads API v1 merged OpenAPI | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/AmazonAdsAPIALLMerged_prod_3p.json | official | 2026-11-09 | v1 `BidStrategy`, goals, KPI, automation flags, recommendation and stream paths |
+| Sponsored Products rule-based bidding overview | https://advertising.amazon.com/API/docs/en-us/guides/rules/bidding-rules/overview | official | 2026-11-09 | eligibility, auto-disable, guardrail guidance, minimum budgets |
+| Getting started with the SP rule-based bidding API | https://advertising.amazon.com/API/docs/en-us/guides/rules/bidding-rules/getting-started | official | 2026-11-09 | endpoints, enums, daily state read |
+| Schedule Bid Rules | https://advertising.amazon.com/API/docs/en-us/guides/rules/bidding-rules/scheduled-bid-rules | official | 2026-11-09 | additive increases, 24 hours, 20 per campaign |
+| Sponsored ads budget rules overview | https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/overview | official | 2026-11-09 | schedule and performance types, 7 and 14 days |
+| Schedule-based rules | https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/schedule-based | official | 2026-11-09 | recurrence, intraday markets, recommended events |
+| Performance-based rules | https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/performance-based | official | 2026-11-09 | metrics by ad type, operators, endpoints |
+| Budget rule evaluation | https://advertising.amazon.com/API/docs/en-us/guides/rules/budget-rules/rule-evaluation | official | 2026-11-09 | start-of-day evaluation, 250 rules, additive, statuses, inconsistent example |
+| Target Promotion | https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/target-promotion | official | 2026-11-09 | Amazon's harvesting |
+| Sponsored Products auto targeting | https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/auto-targeting | official | 2026-11-09 | four expressions, theme bid recommendations |
+| Sponsored Product keyword groups targeting | https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/keyword-group-targeting | official | 2026-11-09 | beta, US, refresh over time |
+| Sponsored Brands theme-based targeting | https://advertising.amazon.com/API/docs/en-us/guides/sponsored-brands/targeting/theme-based-targeting | official | 2026-11-09 | themes, `smartDefault`, bid immutability |
+| Managing Sponsored Brands optimization rules | https://advertising.amazon.com/API/docs/en-us/guides/sponsored-brands/campaigns/managing-optimization-rules | official | 2026-11-09 | ML bids, may add targets, 21 days |
+| Auto-assembled videos | https://advertising.amazon.com/API/docs/en-us/guides/sponsored-products/auto-assembled-videos | official | 2026-11-09 | default opt-in, opt-out field |
+| Prompt Ad Extension reports | https://advertising.amazon.com/API/docs/en-us/guides/reporting/v3/report-types/prompt-ad-extension | official | 2026-11-09 | prompt reports, zero setup |
+| Marketing Stream v2 overview | https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/v2/overview | official | 2026-11-09 | S3 and SQS, Parquet |
+| Marketing Stream v2 datasets overview | https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/v2/datasets/overview | official | 2026-11-09 | two datasets, cadences |
+| Marketing Stream v2 limits | https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/v2/manage-subscriptions/limits | official | 2026-11-09 | destinations, rate, file size |
+| Marketing Stream v2 data delivery | https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/v2/data-delivery | official | 2026-11-09 | totals, versions, at-least-once |
+| Sponsored ads campaign recommendations (beta), Stream v1 | https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/datasets/sponsored-ads-campaign-diagnostics-recommendations | official | 2026-11-09 | v1 recommendation types and triggers |
+| SP budget recommendations and missed opportunities, Stream v1 | https://advertising.amazon.com/API/docs/en-us/guides/amazon-marketing-stream/datasets/budget-recs-missed-opportunities | official | 2026-11-09 | budget recommendation fields |
+| Helium 10 Ads | https://www.helium10.com/tools/advertising/adtomic/ | opinion | 2026-11-09 | features and method claims |
+| Helium 10 pricing | https://www.helium10.com/pricing/ | opinion | 2026-11-09 | MCP included in Diamond |
+| Helium 10 MCP (two blog posts) | https://www.helium10.com/blog/helium-10-mcp/ | opinion | 2026-11-09 | MCP tools and ads workflow |
+| Helium 10 Claude skills | https://www.helium10.com/claude-skills/ | opinion | 2026-11-09 | published skills |
+| Perpetua, What is the ad engine doing? | https://help.perpetua.io/en/articles/5897118-what-is-the-ad-engine-doing | opinion | 2026-11-09 | aggregate ACoS, going-dark mitigation |
+| Perpetua, optimisation for Sponsored Products | https://help.perpetua.io/en/articles/10545202-perpetua-s-optimisation-for-sponsored-products-open-beta | opinion | 2026-11-09 | daily and hourly modes |
+| Perpetua homepage | https://perpetua.io/ | opinion | 2026-11-09 | rank-based rules |
+| Teikametrics, Decoding predictive AI bidding | https://help.teikametrics.com/en/articles/5644036-decoding-predictive-ai-bidding-a-guide-to-optimizing-your-advertising-strategy | opinion | 2026-11-09 | value bidding, discovery phase |
+| Teikametrics Flywheel FAQ and hourly bidding | https://intercom.help/flywheel-20/en/articles/5644025 | opinion | 2026-11-09 | ad-group vs target bids, frequency |
+| Teikametrics Flywheel | https://teikametrics.com/flywheel | opinion | 2026-11-09 | ARI assistant |
+| Scale Insights homepage | https://scaleinsights.com/ | opinion | 2026-11-09 | algorithms, 1% pricing |
+| Scale Insights documentation (llms.txt and about ten pages) | https://docs.scaleinsights.com/llms.txt | opinion | 2026-11-09 | optimum bid, normalisation, windows, negative word rule |
+| m19 | https://www.m19.com/ | opinion | 2026-11-09 | ML automation |
+| m19 vs Perpetua (m19 blog) | https://www.m19.com/blog/m19-vs-perpetua | opinion | 2026-11-09 | m19 and Perpetua pricing as m19 states them |
+| Perpetua Pricing & Plans (added by the verifier) | https://perpetua.io/pricing/ | opinion | 2026-11-09 | $695 base pricing |
+| Quartile | https://www.quartile.com/ | opinion | 2026-11-09 | hourly placements, managed service |
+| Intentwise AI Gateway | https://www.intentwise.com/ai-gateway/ | opinion | 2026-11-09 | MCP for Claude and Copilot |
+| Intentwise, Amazon budget rules | https://www.intentwise.com/blog/amazon-budget-rules | opinion | 2026-11-09 | practitioner reading of budget rules |
+| Ad Badger (home, automate negatives; bid optimization page 404 to the verifier) | https://www.adbadger.com/ | opinion | 2026-11-09 | daily bids, six-month negatives, Claude connection |
+| BidX | https://www.bidx.io/ | opinion | 2026-11-09 | approve-to-execute agents |
+| Adbrew | https://www.adbrew.io/ | opinion | 2026-11-09 | Adbrew MCP, rule and AI automation |
+| Jungle Scout features | https://www.junglescout.com/features/ | opinion | 2026-11-09 | "Automate PPC campaigns" claim, no method stated |
+| GitHub MCP servers and skills (KuudoAI, MarketplaceAdPros, Xnurta, nospicyplease, Deltologic, agentcentral) | https://github.com/ | opinion | 2026-11-09 | community doors |
+| Learning and Trust in Auction Markets | https://arxiv.org/abs/1703.10672 | research | 2026-11-09 | bid recommendations and trust |
+| Autobidders with Budget and ROI Constraints | https://arxiv.org/abs/2301.13306 | research | 2026-11-09 | constrained autobidder theory |
+| Learning to bid with AuctionGym | https://www.amazon.science/publications/learning-to-bid-with-auctiongym | research | 2026-11-09 | counterfactual problem, simulator |
+| amazon-science/auction-gym | https://github.com/amazon-science/auction-gym | research | 2026-11-09 | code is live |
+| AiAds | https://arxiv.org/abs/1907.12118 | research | 2026-11-09 | platform autobidding at scale |
+| PlatformBid | https://arxiv.org/abs/2607.27265 | research | 2026-11-09 | platform-side objectives |
+| Algorithm aversion (Crossref record) | https://doi.org/10.1037/xge0000033 | research | 2026-11-09 | title and venue |
+| Algorithm appreciation (Crossref record) | https://doi.org/10.1016/j.obhdp.2018.12.005 | research | 2026-11-09 | title and venue |
+| Consumer Heterogeneity and Paid Search Effectiveness (Crossref record) | https://doi.org/10.3982/ECTA12423 | research | 2026-11-09 | title and venue |
+| Amazon launches a powerful new generative AI-based selling assistant codenamed Project Amelia (gap fill) | https://www.aboutamazon.com/news/innovation-at-amazon/amazon-project-amelia | official | 2026-11-09 | 2024-09-19 beta in Seller Central; future actions on sellers' behalf |
+| Amazon introduces agentic AI across the seller experience (HTML title "Amazon unveils new agentic AI-powered Seller Assistant ...") (gap fill) | https://www.aboutamazon.com/news/innovation-at-amazon/seller-assistant-agentic-ai | official | 2026-11-09 | 2025-09-17 agentic Seller Assistant; Nova and Claude; Creative Studio for ads; US, free |
+| Amazon Accelerate 2025: A new era of tools and technology for independent sellers (gap fill) | https://sellingpartners.aboutamazon.com/amazon-accelerate-2025-a-new-era-of-tools-and-technology-for-independent-sellers | official | 2026-11-09 | "with permission, take action on behalf of sellers" |
+| Amazon gives sellers an even smarter Seller Assistant and a new plugin for Amazon Quick and Anthropic's Claude (gap fill) | https://www.aboutamazon.com/news/innovation-at-amazon/seller-assistant-plugin-amazon-quick-claude | official | 2026-11-09 | workflows, approval before actions, audit trail, plugin beta, 90% acceptance |
+| At Accelerate 2026, Amazon shares new AI tools as a force multiplier for independent sellers (gap fill) | https://sellingpartners.aboutamazon.com/accelerate-2026-recap | official | 2026-11-09 | drafted SP/SB campaigns; auto-optimization under a seller maximum bid; 40%; 350M Alexa for Shopping users |
+| Amazon launches Canvas Experience for sellers in U.S. and U.K. (gap fill) | https://www.aboutamazon.com/news/innovation-at-amazon/amazon-sellers-canvas-artificial-intelligence | official | 2026-11-09 | canvas on Seller Assistant's architecture; campaign diagnosis; future actions |
+| AI Information — Sell on Amazon (gap fill) | https://sell.amazon.com/ai-information | official | 2026-11-09 | Seller Assistant described as agentic; 90% acceptance attributed to the 2025 Small Business Empowerment Report |
+| Amazon Ads introduces new sponsored ads tools to help small businesses achieve their goals (gap fill) | https://advertising.amazon.com/library/expert-advice/sponsored-ads-small-business-tools | official | 2026-09-22 page, read 2026-11-09 | auto-optimization, one-click switch, US January 2027, 40% and 13%; campaign intelligence October 2026; prompts 48% |
+| Understand delivery alerts and recommendations (help GR6VEAG2EQWG56RD) (gap fill) | https://advertising.amazon.com/help/GR6VEAG2EQWG56RD | official | 2026-11-09 | "Amazon Ads Agent (formerly known as Amazon DSP)" in a DSP passage |
+| Understand Advertiser Accounts (help GXKXR56ZCMJ3WCS4) (gap fill) | https://advertising.amazon.com/help/GXKXR56ZCMJ3WCS4 | official | 2026-11-09 | the same wording for DSP accounts per country |
+| Ads Agent product page, re-opened for footnotes (gap fill) | https://advertising.amazon.com/solutions/products/ads-agent | official | 2026-11-09 | "Based on beta estimates; results are not predictive of future results" |
+| Introducing Amazon Ads Agent, re-opened (gap fill) | https://advertising.amazon.com/library/news/amazon-ads-agent | official | 2026-11-09 | "the new name of the unified platform"; no 67% figure on the page |
+| The 11 biggest takeaways from unBoxed 2026, re-opened (gap fill) | https://advertising.amazon.com/library/news/unboxed-2026-news-announcements | official | 2026-11-09 | 67% and 29% Full-Funnel; 71% and 15% "versus their results before adopting the suite" |
+| Grow total sales with Full-Funnel Campaigns (gap fill) | https://advertising.amazon.com/resources/whats-new/full-funnel-campaigns-from-discovery-to-sales | official | 2026-11-09 | 67% footnote: blended average of other campaigns, Amazon internal June–August 2026 |
+| What agentic shopping means for advertising, re-opened (gap fill) | https://advertising.amazon.com/library/news/agentic-shopping-advertising | official | 2026-11-09 | Rufus and Alexa+ into Alexa for Shopping; Alexa+ devices kept as a separate ad surface |
+| Meet Alexa for Shopping, your personalized, agentic AI assistant on Amazon (gap fill) | https://www.aboutamazon.com/news/retail/alexa-for-shopping-ai-assistant | official | 2026-11-09 | 2026-05-13 merger; US app, website and Echo Show |
+| An author's guide to Sponsored Products, re-opened (gap fill) | https://advertising.amazon.com/library/guides/authors-guide-to-sponsored-products | official | 2026-11-09 | 100% top of the first page, 50% elsewhere |
+| Guide to dynamic bidding and help GCU2BUWJH2W3A8Z7, re-opened (gap fill) | https://advertising.amazon.com/library/guides/dynamic-bidding-sponsored-products | official | 2026-11-09 | "Currently ... up to 100% ... for all placements" |
+| Amazon Ads reporting overview; Reporting API v1 overview; Limits and best practices; Stream v2 Overview (gap fill) | https://d3a0d0y2hgofx6.cloudfront.net/en-us/guides/reporting/overview.md (and `ads-v1/overview.md`, `ads-v1/limits.md`, `amazon-marketing-stream/v2/overview.md`) | official | 2026-11-09 | concurrency per advertiser (3 pages) and per client id (2 pages) |
+| Amazon Ads API v1 merged OpenAPI, re-read (gap fill) | https://d1y2lf8k3vrkfu.cloudfront.net/openapi/en-us/dest/AmazonAdsAPIALLMerged_prod_3p.json | official | 2026-11-09 | no auto-optimization or maximum-bid field |
+| Pacvue Launches Pacvue Agent (press release, Destination CRM) (gap fill) | https://www.destinationcrm.com/Articles/CRM-News/CRM-Across-the-Wire/Pacvue-Launches-Pacvue-Agent-174367.aspx | opinion | 2026-11-09 | approval-based execution, guardrails, AMC SQL |
+| Pacvue Launches Pacvue Prism (press release, WBOC) (gap fill) | https://www.wboc.com/online_features/press_releases/pacvue-launches-pacvue-prism-powering-the-industrys-first-agentic-commerce-grid/article_e81a6b7d-95a8-5dcb-8bbd-97da57251b7b.html | opinion | 2026-11-09 | Agentic Commerce Grid, MCP |
+| PPC Land, Pacvue Agent promises 200x faster commerce media workflows (Luis Rijo) (gap fill) | https://ppc.land/pacvue-agent-promises-200x-faster-commerce-media-workflows/ | opinion | 2026-11-09 | Amazon-only at launch; governed execution; figures |
+| PPC Land, Pacvue Prism launches today (Luis Rijo) (gap fill) | https://ppc.land/pacvue-prism-launches-today-to-kill-the-channel-budget-debate/ | opinion | 2026-11-09 | halo regression model; MCP server; figures without methodology |
+| Pacvue case studies on casestudies.com: beverage manufacturer; OLLY (gap fill) | https://www.casestudies.com/company/pacvue/ | opinion | 2026-11-09 | Dynamic Dayparting on Stream, 14-day patterns, bids up and down by hour |
+| Skai Retail Media Solutions (gap fill) | https://skai.io/retail-media/ | opinion | 2026-11-09 | Budget Navigator, automated negatives, AI dayparting, automated actions, harvesting, SOV |
+| Skai marketing measurement (Impact Navigator) (gap fill) | https://skai.io/impact-navigator/ | opinion | 2026-11-09 | incrementality testing on aggregate data |
+| Skai, Retail Media Incrementality: Skai's Innovative Approach (gap fill) | https://skai.io/blog/retail-media-incrementality-skais-innovative-approach/ | opinion | 2026-11-09 | holdout as gold standard; branded leakage 5–15% |
+| Skai, Unlocking Incrementality on Amazon with AMC via Skai (gap fill) | https://skai.io/blog/unlocking-incrementality-on-amazon-with-amazon-marketing-cloud-via-skai/ | opinion | 2026-11-09 | AMC route |
+| Skai Launches Agent Connect (press release) and Skai homepage (gap fill) | https://skai.io/press-releases/skai-launches-agent-connect-opening-its-omnichannel-marketing-infrastructure-to-external-ai-agents/ | opinion | 2026-11-09 | outside agents edit bids and budgets with human approval |
+| My Amazon Guy, Amazon Accelerate 2026 Announcements: New or Recycled? (Noah Wickham) (gap fill) | https://myamazonguy.com/news/amazon-accelerate-2026-announcements/ | opinion | 2026-11-09 | Seller Assistant live; keyword auto optimization 40% "in early results" |
+| Improvado, Amazon unBoxed 2026 (Ivo Petrovic) (gap fill) | https://improvado.io/blog/amazon-unboxed-2026-retail-media-measurement | opinion | 2026-11-09 | Full-Funnel figure "not a holdout test on your account" |
+| AiAds full text (gap fill) | https://arxiv.org/pdf/1907.12118 | research | 2026-11-09 | non-random groups; tables 2–3; "will gradually abate" |
+| Limiting Bias from Test-Control Interference in Online Marketplace Experiments (Holtz and Aral) (gap fill) | https://arxiv.org/abs/2004.12162 | research | 2026-11-09 | Blake and Coey's factor of 2; Fradkin's over 100% |
+| Why marketplace experimentation is harder than it seems (Blake and Coey), Crossref record (gap fill) | https://doi.org/10.1145/2600057.2602837 | research | 2026-11-09 | title, venue, pages |
+| Regression to the mean: what it is and how to deal with it (Barnett, van der Pols, Dobson), PubMed abstract (gap fill) | https://pubmed.ncbi.nlm.nih.gov/15333621/ | research | 2026-11-09 | definition; selection on a baseline value |
+
+Unreachable, with reasons:
+
+- Helium 10 knowledge base articles (kb.helium10.com, including *How Do I Use Automation Within Adtomic*, *Budget Manager*, *Keyword Tracker Based Bid Automation Rules*): 403 to curl and the fetch tool. Snippets only.
+- Pacvue (pacvue.com, the Pacvue for Amazon page, blog posts on bid, placement and dayparting rules, customer stories): 403 (Cloudflare challenge) to curl and the fetch tool, again in the gap fill. Pacvue's own words were read on wire releases and casestudies.com instead; its help centre is not public.
+- The Internet Archive (web.archive.org): its availability API listed Pacvue snapshots, but the snapshot host refused the TLS connection to curl and the fetch tool refused the host.
+- Skai's help centre: not public; Skai's Amazon page still returns image bytes.
+- Blake and Coey on dl.acm.org: a browser challenge; the paper was not read, only its Crossref record and Holtz and Aral's summary.
+- A Seller Central help page on Seller Assistant or on auto-optimization: none found by search; `sell.amazon.com/tools/seller-assistant` and `sell.amazon.com/seller-assistant` answered 404.
+- Sellozo (home, features, pricing): a JavaScript shell with no readable text.
+- Intentwise help article on harvesting with automated rules: 404.
+- Perpetua's /amazon/ page: 404. (Perpetua pricing, unreachable to the research seat, was read by the verifier.)
+- Skai's Amazon page (skai.io/retail-media/amazon/): returned image bytes (still so to the verifier).
+- m19 pricing page: answered "in progress"; its figures were read on m19's own comparison post instead.
+- Ad Badger *Bid optimization* (https://www.adbadger.com/bid-optimization/): 404 to the verifier.
+- Amazon's unBoxed 2025 Ads Agent what's-new page: 404 (the news article was read instead).
+- `https://advertising.amazon.com/ads-agent`: redirects to Amazon sign-in; not followed.
+- Amazon Ads MCP server endpoints (`advertising-ai*.amazon.com/mcp`, `/mcp/lite`) and the skills download (`advertising-api*.amazon.com/skills`): forbidden hosts for this seat; not called.
+- Amazon Ads help pages (for example GDLQ5D2BNFCAU3TW on bidding rules) render as a JavaScript shell to the fetch tool; the verifier read GDLQ5D2BNFCAU3TW and GCU2BUWJH2W3A8Z7 from the page content embedded in the plain-GET HTML (`window.__scV2Content`), so they are no longer unreachable.
+- The web search budget for the session ran out during the published-research pass; papers on autobidding by Aggarwal, Badanidiyuru and Mehta, and Balseiro and co-authors, were not located and are not cited.
+
+## Open questions
+
+1. The MCP open-beta note (2026-02-18) names a workflow to "automate keyword harvesting from automatic to manual campaigns"; the current `campaign_management.json` and the skills library show no such tool or skill. Was it withdrawn, renamed, or moved behind the `/skills` download the seat did not call?
+2. The v2 optimization-rule model (`application/vnd.spoptimizationrules.v2+json`, with `BUDGET` and `TARGETING` categories and performance conditions) has no guide page. Is it live for all advertisers, and does a `TARGETING` rule do anything other than raise a bid, given that the only action is `ADOPT` / `INCREMENT`?
+3. Ads Agent's proposals and approvals have no API page. The change history API states it "doesn't return information about who made the change", so a change applied through Ads Agent is not distinguishable there; is there any other read that marks Amazon-initiated or agent-applied changes?
+4. After Stream v1 stops, the Recommendations API serves four types. Will the consolidated SP recommendations, the budget recommendations and the v1 diagnostics types move into it, or stay on v3 endpoints that are not on the deprecation list?
+5. The Recommendations API guide writes `GET /adsApi/v1/list/recommendations`; the merged model lists `GET /adsApi/v1/recommendations` and SB-specific create and query-types paths. Which path answers?
+6. Partner Opportunities are written for partners. Does a direct advertiser's application receive them?
+7. Can the door's caller fetch `GET /skills?name=all&download=true` on the Ads API origin, and do Amazon's terms allow a third party to reuse those skills?
+8. The budget rule evaluation example contradicts its own additive rule (the 20% rule with ROAS > 3 "not applied" at ROAS 4). Which reading holds at the edge?
+9. The release notes date the Review Requests entry November 16, 2026, after the reading date. Is that a scheduled release or a typing slip?
+10. Rufus became part of Alexa for Shopping in May 2026, and existing sponsored ads campaigns are automatically eligible there. Does the 2024 line "Your campaign reports will not include Rufus metrics" still hold for Alexa for Shopping, are its clicks and spend in campaign totals with no placement label, and is there any campaign-level way to keep a campaign out of it?
+11. Auto-assembled videos: are their impressions and spend broken out in Reporting v1 by `videoType`, so their effect can be read?
+12. The 100% "up and down" figure on four Amazon pages against the 100%/50% split on the author's guide: which is live in each marketplace, and since when? No Amazon page dates a change; `areas/bids-and-placements.md` owns the question.
+13. Helium 10 and Pacvue method figures (default target ACoS, threshold windows) and Perpetua's "lower the target by no more than 10%" were read only from search snippets; they stand unverified. (Perpetua's 1.3× ceiling was verified on its help page as a recommendation.)
+14. No published study measures Amazon's dynamic bidding or a third-party Amazon bidder's effect on seller profit; this is a gap in the literature, not in the seat's search alone.
+15. The console help page on bidding rules lets Amazon set the ROAS target when the advertiser gives none and states no 10-day, 10-conversion floor, while the API guide states both. Which eligibility governs an API-created rule?
+16. The Sponsored Products Amazon-built audiences for bid boosting were announced as a "Preview" on 2025-03-12; no later release note announces general availability. Are they generally available?
+17. Auto-optimization (US, January 2027 per Amazon Ads; announced inside Seller Assistant at Accelerate 2026): which campaign field carries the switch and the seller's maximum bid, whether that maximum is per target or per campaign, what counts as "performing" and "underperforming", and whether it is the same feature Ads Agent's product page calls "auto-optimized Sponsored Products campaigns".
+18. Seller Assistant's autonomy is described two ways on Amazon's pages ("acts on its own" against "nothing happening without approval"). For auto-optimization the Amazon Ads page describes changes made on the advertiser's behalf and reversible afterwards. Which applies to a drafted campaign once auto-optimization is on?
+19. Do campaigns drafted by Seller Assistant carry any marker (name pattern, `autoCreationSettings`, a tag) that an API reader can see?
+20. Are the three "40% faster first sale" figures (Ads Agent product page, Accelerate 2026 recap, small-business tools page) one beta measured once, and what was the comparison group?
+21. What does Pacvue's or Skai's automation do by default on a newly connected Amazon account? Neither public site says; both help centres are closed.
+
+## Positions
+
+**P-automation-and-amazon-tools-1.** Amazon changes a seller's bids, budgets, targets and creative through switches on the campaign, portfolio or ad extension (and, for Alexa for Shopping, Alexa+ and prompts, through automatic eligibility with no documented campaign switch), so an account's results are not explained by the seller's own edits alone. Sources: [official] SP 3.0 OpenAPI; Amazon Ads API v1 merged OpenAPI; *Sponsored Products auto targeting*; *Auto-assembled videos*; *Managing Sponsored Brands optimization rules*; release notes 2025-07-19. Tags: bidding, budgets, targeting, creative, diagnosis.
+
+**P-automation-and-amazon-tools-2.** Under "dynamic bids - up and down" Amazon's current help, library guide and API models say Amazon may raise or lower a bid by up to 100% for all placements ("Ensure you are prepared to potentially spend double"), while Amazon's author's guide still says up to 100% for top of the first page and up to 50% for other placements; `areas/bids-and-placements.md` P-8 owns this fact. Sources: [official] *Bidding strategies for Sponsored Products*; *Guide to dynamic bidding - up and down with Sponsored Products*; SP 3.0 OpenAPI `BiddingStrategy`; v1 merged OpenAPI `BidStrategy`; *An author's guide to Sponsored Products*. Tags: bidding, placements.
+
+**P-automation-and-amazon-tools-3.** Sponsored Products schedule and performance optimization rules and sponsored ads budget rules can only add a percentage to a bid or a budget, while rule-based bidding and Sponsored Brands optimization rules let Amazon lower bids as well as raise them; no Amazon rule pauses a target, and a decrease or pause the seller chooses is a direct write (Amazon's announced auto-optimization, section L, is not a rule and does pause keywords). Sources: [official] SP 3.0 OpenAPI optimization-rule schemas; *Schedule Bid Rules*; *Sponsored ads budget rules overview*; *Budget rule evaluation*; *Bidding rules for Sponsored Products*; *Managing Sponsored Brands optimization rules*. Tags: rules, bidding, budgets.
+
+**P-automation-and-amazon-tools-4.** Rule-based bidding holds a campaign-level ROAS guardrail that Amazon does not guarantee, may raise a bid up to 5 times the adjusted bid, puts any schedule bid rules on hold, and is switched off by Amazon, restoring the earlier strategy and bids, when the campaign's recent ROAS collapses against its earlier level. Sources: [official] *Sponsored Products rule-based bidding overview*; *Getting started with the Sponsored Products rule-based bidding API*; *Bidding rules for Sponsored Products*. Tags: bidding, rules.
+
+**P-automation-and-amazon-tools-5.** When several schedule bid rules or several budget rules hold at once, Amazon adds their increases together, and schedule bid rules sit on top of placement adjustments and the bidding strategy. Sources: [official] *Schedule Bid Rules*; *Budget rule evaluation*. Tags: rules, bidding, budgets, seasons.
+
+**P-automation-and-amazon-tools-6.** With unspent-budget sharing on, a campaign in a portfolio can spend beyond its own daily budget, up to twice it, from budget its sibling campaigns left unspent earlier in the month. Sources: [official] Ads API release notes, 2025-07-19. Tags: budgets, portfolios.
+
+**P-automation-and-amazon-tools-7.** Amazon's bid, budget and targeting recommendations are dated estimates refreshed about daily, and applying one is the same write to the named campaign, keyword or placement that any other change uses. Sources: [official] SP 3.0 OpenAPI Consolidated Recommendations and budget recommendation descriptions; *Recommendations API overview*; *Partner Opportunities - Campaign Optimization*. Tags: recommendations.
+
+**P-automation-and-amazon-tools-8.** Amazon's cross-product Recommendations API serves the types its *Recommendation types* page lists (four on 2026-11-09, among them campaigns paused by the system and a free-trial expansion of Sponsored Products campaigns into other countries), Amazon adds types over time, and its guide tells readers to tolerate `recommendationType` values they do not recognise. Sources: [official] *Recommendation types*; *How to call the API*; *Recommendations API overview*. Tags: recommendations, global, diagnosis.
+
+**P-automation-and-amazon-tools-9.** Amazon runs its own harvesting: target promotion groups move converting search terms and ASINs from an auto ad group into linked manual keyword and product ad groups, on Amazon's reading of their historical performance. Sources: [official] *Target Promotion*; SP 3.0 OpenAPI `TargetPromotionGroups`; *Partner Opportunities - Campaign Optimization*. Tags: harvesting, search terms, campaign structure.
+
+**P-automation-and-amazon-tools-10.** Several Amazon targeting modes choose and refresh targets themselves: automatic targeting, keyword groups, Sponsored Brands theme targeting, Sponsored Brands cost-control rules that may add targets, and the v1 `autoCreateTargets` flag. Sources: [official] *Sponsored Products auto targeting*; *Sponsored Product keyword groups targeting*; *Sponsored Brands theme-based targeting*; *Managing Sponsored Brands optimization rules*; v1 merged OpenAPI `AutoCreationSettings`. Tags: targeting.
+
+**P-automation-and-amazon-tools-11.** In the US every Sponsored Products campaign is opted in to auto-assembled videos, which Amazon assembles from existing uploaded video clips, unless the seller opts out on the campaign, and the opt-out leaves videos already delivering running until they are paused through the ad extension. Sources: [official] *Auto-assembled videos*; release notes 2026-08-09. Tags: creative, Sponsored Products.
+
+**P-automation-and-amazon-tools-12.** In May 2026 Amazon brought Rufus and Alexa+ together as Alexa for Shopping on the US Amazon Shopping app, website and Echo Show, while Alexa+ continues on Alexa devices and Alexa.com; existing sponsored ads campaigns are automatically eligible to serve in Alexa for Shopping and to be optimized as Sponsored Products and Sponsored Brands prompts, and are automatically enrolled in Alexa+ Conversational Ads on Alexa-enabled devices; no campaign switch for these surfaces is documented, and individual prompts can be paused in the console. Sources: [official] *Meet Alexa for Shopping, your personalized, agentic AI assistant on Amazon* (2026-05-13); *What agentic shopping means for advertising* (2026-06-11); *Prompt Ad Extension reports*; release notes 2024-09-05 and 2025-07-15. Tags: placements, reporting, creative.
+
+**P-automation-and-amazon-tools-13.** Amazon Ads Agent is, from 2026-09-29, "the new name of the unified platform" that holds sponsored ads, Display, Video, and Audio, and Full-Funnel Campaigns; inside it, the conversational experience is a closed beta that proposes keyword, bid and budget changes and applies them after the advertiser approves, with no public API; the help centre's "Amazon Ads Agent (formerly known as Amazon DSP)" is the same platform name written into passages about DSP settings and accounts. Sources: [official] *Introducing Amazon Ads Agent*; *Ads Agent: An AI marketing assistant from Amazon Ads*; *The 11 biggest takeaways from unBoxed 2026*; *Understand delivery alerts and recommendations* (help GR6VEAG2EQWG56RD); *Understand Advertiser Accounts* (help GXKXR56ZCMJ3WCS4). Tags: Ads Agent, recommendations.
+
+**P-automation-and-amazon-tools-14.** Full-Funnel Campaigns and DVA+ take a goal, products, creative and budget and leave planning, targeting and optimisation to Amazon, with advanced settings for advertisers who keep control. Sources: [official] *Introducing Amazon Ads Agent*; *The 11 biggest takeaways from unBoxed 2026*. Tags: ad products, Ads Agent.
+
+**P-automation-and-amazon-tools-15.** Amazon's creative tools (Creative Agent, Video Generator, one-click image generation) produce ad creative from the seller's product pages and assets at no added cost, with availability set by country, and Sponsored Brands headline optimization, with the advertiser's consent, tests headlines and serves the best performer. Sources: [official] *Transform campaign and asset development with Creative Agent*; *Amazon Ads launches new agentic AI tool that creates professional-quality ads*; *Video Generator is now available for Sponsored Brands video campaigns in nine marketplaces*; *AI creative solutions*; release notes 2025-04-17. Tags: creative.
+
+**P-automation-and-amazon-tools-16.** The Amazon Ads MCP server is a host separate from the Ads API, with deterministic tools over the same campaign, report and account objects, and Amazon's published MCP skills cover reporting, accounts, eligibility, AMC and DSP campaign creation, not Sponsored Products bid or budget optimisation. Sources: [official] *Overview* (Amazon Ads MCP Server); *Connecting to the Amazon Ads MCP Server*; MCP tool specifications; MCP skills library pages. Tags: MCP, skills.
+
+**P-automation-and-amazon-tools-17.** Marketing Stream v2 delivers hourly sponsored-ads performance to an S3 bucket and SQS queue the advertiser owns, and Stream v1's recommendation and messaging datasets end on 2027-06-30 without a v2 equivalent. Sources: [official] Marketing Stream v2 pages; *Deprecations*; Stream v1 dataset pages. Tags: data, reporting.
+
+**P-automation-and-amazon-tools-18.** Third-party Amazon bid tools set each target's bid from its value per click times a target ACoS, on a daily or hourly schedule, and write it through the same Ads API endpoints. Sources: [opinion] Scale Insights documentation; Teikametrics help; Perpetua help; Helium 10 *Helium 10 Ads*; Ad Badger. Tags: bidding, third-party tools.
+
+**P-automation-and-amazon-tools-19.** Third-party tools state that they add negatives from wasted spend, hour-of-day bid decreases and pauses, pausing and reviving targets, and bids tied to organic rank, none of which Amazon's documented rules do; harvesting from auto to manual targeting is offered both by these tools and by Amazon's target promotion groups, and Amazon's announced auto-optimization (US, January 2027) adds keywords and pauses underperforming ones. Sources: [official] SP 3.0 OpenAPI optimization-rule schemas; *Target Promotion*; [opinion] Helium 10 *Helium 10 Ads*; Scale Insights *Overview of Each Automation Algorithm*; Perpetua homepage; Ad Badger. Tags: third-party tools, harvesting, negatives, dayparting.
+
+**P-automation-and-amazon-tools-20.** Helium 10, Intentwise, Ad Badger and Adbrew now sell MCP connections or Claude skills over a seller's advertising data, alongside Amazon's own MCP server. Sources: [official] *Overview* (Amazon Ads MCP Server); [opinion] Helium 10 pricing and *Claude skills*; Intentwise *AI Gateway*; Ad Badger; Adbrew. Tags: MCP, third-party tools.
+
+**P-automation-and-amazon-tools-21.** A report shows what happened at the bid that ran, never what a target would have done at another bid. Sources: [research] Jeunen, Murphy and Allison, *Learning to bid with AuctionGym*. Tags: bidding, measurement.
+
+**P-automation-and-amazon-tools-22.** Amazon's change history API (`POST /history`) returns changes to Sponsored Products and Sponsored Brands campaigns, not Sponsored Display, and "doesn't return information about who made the change". Sources: [official] Change history OpenAPI. Tags: diagnosis, measurement.
+
+**P-automation-and-amazon-tools-23.** Seller Central's Seller Assistant (Project Amelia in its 2024 beta, agentic from September 2025) checks whether each product's listing and inventory are ready to advertise and drafts Sponsored Products and Sponsored Brands campaigns that the seller approves "in a keystroke" or opens in the ad console. Sources: [official] *At Accelerate 2026, Amazon shares new AI tools as a force multiplier for independent sellers*; *Amazon launches a powerful new generative AI-based selling assistant codenamed Project Amelia*; *Amazon introduces agentic AI across the seller experience*. Tags: Seller Assistant, launch, campaign structure.
+
+**P-automation-and-amazon-tools-24.** Amazon describes an SP auto-optimization that, once switched on with one click in campaign settings, "expands product targeting", "adds relevant keywords and pauses underperforming ones" and "identifies seasonal opportunities" within the campaign's existing settings, holds "every bid under a maximum the seller sets", and lets the advertiser "edit or reverse actions at any time"; Amazon Ads dates its US availability to January 2027, and no page states its thresholds. Sources: [official] *Amazon Ads introduces new sponsored ads tools to help small businesses achieve their goals* (2026-09-22); *At Accelerate 2026, Amazon shares new AI tools ...* (2026-09-29). Tags: Seller Assistant, harvesting, targeting, bidding.
+
+**P-automation-and-amazon-tools-25.** On 2026-11-09 no Ads API release note or model field documents auto-optimization or Seller Assistant: the v1 campaign carries `autoCreationSettings` (`autoCreateTargets`, `autoManageCampaign`) and no auto-optimization or maximum-bid field, and the change history API does not record who made a change. Sources: [official] Ads API release notes; Amazon Ads API v1 merged OpenAPI (`Campaign`, `AutoCreationSettings`); Change history OpenAPI. Tags: Seller Assistant, API, diagnosis.
+
+**P-automation-and-amazon-tools-26.** Seller Assistant workflows run "even when the seller isn't logged in", and for each workflow the seller chooses "whether it just surfaces recommendations or takes actions", with actions reviewed and approved before they are carried out and logged in an audit trail; Amazon's examples are pricing, restocking, listings and account health, not ads. Sources: [official] *Amazon gives sellers an even smarter Seller Assistant and a new plugin for Amazon Quick and Anthropic's Claude*; *At Accelerate 2026, Amazon shares new AI tools ...*. Tags: Seller Assistant.
+
+**P-automation-and-amazon-tools-27.** Amazon's published lift figures for its own automation are vendor-reported comparisons from beta or internal data, with no randomised holdout described: the Ads Agent figures are "Based on beta estimates; results are not predictive of future results", Full-Funnel's 67% is "against the blended average of other campaign performance across beta advertisers", the creative-tool gains are "versus their results before adopting the suite", and the auto-optimization, ads data manager and prompt figures compare adopters or clickers with those who did not. Sources: [official] *Ads Agent: An AI marketing assistant from Amazon Ads*; *Grow total sales with Full-Funnel Campaigns*; *The 11 biggest takeaways from unBoxed 2026*; *Amazon Ads introduces new sponsored ads tools ...*; release notes 2025-04-17; [opinion] Improvado (Ivo Petrovic). Tags: measurement, Ads Agent, Full-Funnel.
+
+**P-automation-and-amazon-tools-28.** Rule-based bidding through the API is open only to campaigns running at least 10 days with at least 10 conversions in the last 30, so the campaigns on it had converted before it was switched on, and a comparison with ineligible campaigns mixes the rule's effect with that gate. Sources: [official] *Sponsored Products rule-based bidding overview*. Tags: bidding, rules, measurement.
+
+**P-automation-and-amazon-tools-29.** When units are chosen on a baseline value, later measurements move toward the mean on their own, so a campaign handed to automation after an unusually good or bad period shows a change that is partly regression to the mean. Sources: [research] Barnett, van der Pols and Dobson, *Regression to the mean: what it is and how to deal with it*, International Journal of Epidemiology (2005). Tags: measurement.
+
+**P-automation-and-amazon-tools-30.** In an auction, a gain measured on a few early adopters can overstate the gain once many advertisers use the same automation: Baidu's AiAds authors expected its advantage to "gradually abate" as more advertisers adopted it, and Blake and Coey found, in an eBay email-marketing experiment, that naive estimates ignoring test-control interference "exaggerate the treatment's effectiveness by about a factor of 2". Sources: [research] Yang et al., *AiAds*, KDD 2019; Blake and Coey, ACM EC 2014, as summarised by Holtz and Aral 2020. Tags: measurement, bidding.
+
+**P-automation-and-amazon-tools-31.** Amazon's pages state Reporting v1's one-report concurrency both per advertiser (the GA note, the reporting overview, the Stream v2 overview) and per client id (the v1 overview, *Limits and best practices*, concurrency "as low as 1"); `method/reporting.md` P-reporting-15 owns the fact. Sources: [official] those five pages. Tags: reporting, API.
+
+**P-automation-and-amazon-tools-32.** Pacvue states that Pacvue Agent turns recommendations into "approval-based execution with defined guardrails" for Amazon Ads, and its case studies describe Dynamic Dayparting that raises bids in high-conversion hours and lowers them in low ones from Amazon Marketing Stream data and 14-day hourly patterns. Sources: [opinion] Pacvue press release (*Pacvue Launches Pacvue Agent*); Pacvue case studies on casestudies.com; PPC Land (Luis Rijo). Tags: third-party tools, dayparting.
+
+**P-automation-and-amazon-tools-33.** Skai states that its platform makes daily bid and budget optimizations (Portfolios & Budget Navigator), automates negative keywords and keyword harvesting on user conditions, runs AI dayparting from Amazon's hourly Stream data, sells incrementality testing on aggregate data, and lets outside agents change bids and budgets with every change subject to human approval. Sources: [opinion] Skai *Retail Media Solutions*; Skai marketing measurement page (Impact Navigator); Skai press release *Skai Launches Agent Connect*. Tags: third-party tools, measurement.
+
+### Seat's positions (enter no skill)
+
+- [seat] Every write a third-party tool makes to Amazon is reachable through the open door; what a tool has that cx lacks is a clock that runs without being asked and its own store of history.
+- [seat] Amazon's rule engines (raise-only schedule and budget rules, and rule-based bidding, which moves bids both ways) are the part of a tool's schedule that cx can hand to Amazon to run between conversations.
+- [seat] Because Amazon raises bids, shares budgets, adds targets and adds creative on its own, a before/after reading of any manual change starts from the list of switches in section A.
+- [seat] The "AI proposes, the advertiser approves" pattern of Ads Agent and BidX is a product choice of those vendors; for cx, writes carry no switch of ours and the risk is the seller's.
+- [seat] The platform-side objective in PlatformBid and the "spend double" warning are reasons a seller's value per click, not Amazon's suggested bid, is the reference for a bid.
+- [seat] Because auto-optimization and Seller Assistant drafts leave no API marker found on 2026-11-09, targets that appear or pause on a campaign without a seller's edit may come from either of them, from `autoCreateTargets`, from a keyword group or from a third-party tool; section A is the list to read against.
+- [seat] An Amazon beta figure is the average over the campaigns Amazon or the advertisers chose to put in the beta, against a comparison group of Amazon's choosing; for one seller's campaign the expected effect is unknown until its own change is read (`method/change-measurement.md`).
+
+## Verification
+
+Verified 2026-11-09 by an adversarial verifier seat.
+
+Method. Every cited page was re-opened on 2026-11-09 by plain GET: the 34 Ads API guide, MCP and release-note pages as their cloudfront markdown (`https://d3a0d0y2hgofx6.cloudfront.net/en-us/<path>.md`), with the frontmatter title and H1 checked against the page name used here; the SP 3.0, v1 merged, SB 3.0 and change history OpenAPI models parsed for every enum, `maxItems`, required list, media type and path quoted; the 15 MCP tool specifications parsed for tool names and descriptions, and the skills library located through `toc2.json`; the Amazon Ads news, product, what's-new and creative pages fetched as HTML and reduced to text; two console help pages read from the content embedded in their HTML; the amzn/ads-advanced-tools-docs repository and five community repositories through GitHub's public API; the four arXiv papers through the arXiv export API, the AuctionGym page on amazon.science, and the three DOIs through Crossref; and every vendor page re-fetched. Quotes were compared verbatim. Searches for 2025–2026 changes found the May 2026 merger of Rufus and Alexa+ into Alexa for Shopping, which the file had missed. No position was worded as behaviour (no ask-list, sequence, threshold-as-rule, ceiling or talk rule); positions that misstated the facts were rewritten.
+
+Corrections (was → now):
+
+1. Section B and Stable 3, P-3: "every Amazon bid rule can only raise a bid" / "no Amazon rule lowers a bid ... on performance" → SP optimization rules and budget rules only raise; rule-based bidding ("increasing or decreasing them as needed") and SB optimization rules ("if costs are too high, bids are reduced") lower bids too (*Bidding rules for Sponsored Products*; *Managing Sponsored Brands optimization rules*; *Sponsored Products rule-based bidding overview*).
+2. Section B: the 5× cap and the exclusion with schedule rules, "as read by the sibling file; not re-opened" → verified on help GDLQ5D2BNFCAU3TW, quoted, with the console's differing eligibility (Amazon-set ROAS target, "wait at least two weeks") added; Stable 4 and P-4 extended accordingly.
+3. Section B: v2 optimization rule "carries 1 to 10 conditions ... exactly one targeting block" → those fields are optional; required are `ruleCategory`, `ruleName`, `ruleSubCategory`, `status`.
+4. Section B: SP Amazon-built audiences (2025-03-12) stated as available → the entry is a "Preview" with no later GA note; SB NTB worldwide date 2025-03-24 added.
+5. Section B and "What cx must read": `GET /sp/campaigns/{id}/optimizationRules` → no such operation in the SP 3.0 model; rules are found with `POST /sp/rules/optimization/search`.
+6. Section D: consolidated recommendations described with one media type → `GET` answers the v1 shape, `POST` answers a different v2 shape (`recommendationType`, `forecastEstimates`, `nextToken`).
+7. Section D: Recommendations API quotes "across ad products" and "insights explaining why" attributed to the overview → the first is the release note's wording; the overview says "across Amazon advertiser campaigns" and "insights related to why the customer is receiving the recommendation". Pagination absence and the 5 rps limit added.
+8. Section D: `POST /sb/recommendation/optimization` → the SB 3.0 model and guide write `/sb/recommendations/optimization`; the release note's singular form recorded as such.
+9. Section D: Partner Opportunities 29% figure undated → footnote "Amazon internal data, WW, September 2022 – August 2023" added.
+10. Section F, Stable 10, P-11: auto-assembled videos made "from existing product images", "All Sponsored Products campaigns are opted in", "Released 2026-08-09" → "assembled from existing uploaded video clips", "All campaigns are opted in to auto-assembled videos by default", and 2026-08-09 is the documentation date.
+11. Section F: Creative Agent quote "agentic AI-powered creative at no cost" → not on the page; replaced with the verbatim "produce high-quality advertising content quickly and at no cost" and the page's own availability lists; page name corrected.
+12. Section F: creative tool quotes "conducts research, generates concepts, creates storyboards..." and Jay Richman's "handles everything from research and ideation through final production" → neither is on the page; replaced with the page's verbatim sentences and Richman's actual quote; page name corrected to *Amazon Ads launches new agentic AI tool that creates professional-quality ads*.
+13. Section F: Rufus and Alexa+ as separate surfaces with "Rufus metrics" excluded → Rufus and Alexa+ were merged into Alexa for Shopping in May 2026; existing campaigns are automatically eligible there and as prompts, and Alexa+ Conversational Ads enrol them automatically (*What agentic shopping means for advertising*, 2026-06-11); the 2024 and 2025 notes kept as dated history; table row A, the scope line, P-1, P-12 and Stable 16 updated.
+14. Section F: generative AI page listing "live image features", named *Generative AI ad solutions* → no live-image feature on the page; name corrected to *AI creative solutions*; SB video country list added.
+15. Section F: headline optimization scope added ("on Product Collection or Store Spotlight creatives") with the Product Collection deprecation (January 2027).
+16. Section G: three campaign types "(sponsored ads, DVA+, Full-Funnel)" → Amazon's wording "sponsored ads, Full-Funnel Campaigns, and Display, Video, and Audio"; "25% additional unique customers on DVA campaigns" attributed to the right page with its footnote.
+17. Section G: unBoxed 2025 quote "takes actions at the direction of advertisers" → verbatim "actions at the direction of advertisers" (sic); page name *unBoxed 2025: Simplifying advertising through innovation*.
+18. Section H/I and all citations: MCP page names *Amazon Ads MCP Server overview*, *Get started with the Amazon Ads MCP Server*, *Get started with Amazon Ads MCP Server Lite*, *MCP skills overview*, *Get started with skills* → the pages' own titles (*Overview*, *Connecting to the Amazon Ads MCP Server*, *... (Lite)*, *Overview*, *Download Skills from the Amazon Ads MCP Server*).
+19. Section H: Dynamic mode "the model picks accounts" → "the LLM asks you to provide the account identifiers"; `disabledTools` is a Kiro client setting; "Every tool input carries ... `skill`" → 28 of 32 (not the portfolio tools); "user management" domain → `users`, `user_invitations`, `user_permissions`, `user_roles` (no `user_management.json`), plus advertiser product group eligibility.
+20. "What cx must read" and the door table: the change history API as the way to tell seller, tool or Amazon changes apart → the model states it "doesn't return information about who made the change" and covers SP and SB only; P-22 and Stable 17 added; open question 3 narrowed.
+21. P-8: "returns recommendation types beyond any fixed list" → it serves the types its page lists (four today); others return `INVALID_RECOMMENDATION_TYPE`; readers are told to tolerate new types.
+22. P-19: harvesting listed as something Amazon's rules do not do → harvesting is also Amazon's (target promotion groups); the position now lists only what Amazon's documented rules do not do.
+23. Helium 10: "(a Pacvue company) renamed Adtomic" → neither fact is on the pages read; claims now quoted verbatim; the Listing Analyzer skill and the "2% management fee" added.
+24. Perpetua: 1.3× ceiling "from search snippets" → verified on the help page as a recommendation ("we recommend 1.3* Target ACoS"); pricing "not reached" → read on *Perpetua Pricing & Plans* ($695).
+25. Scale Insights: "default analysis window is 60 days skipping the last 2" → not on the page; the window is user-set (example 2 ignored, 7 analysed; up to 180, 365 for the Bidding Rule); algorithm list completed (Whitelist, Default Bidding, Status); two further bid formulas noted.
+26. m19 pricing "search snippet" → read on m19's comparison post; Quartile "by machine learning" → "based on real-time data"; Teikametrics and Ad Badger claims replaced by verbatim quotes.
+27. Jungle Scout: "no PPC automation" → the page claims "Automate PPC campaigns", with no method stated.
+28. Dynamic bidding seat note: "three current Amazon pages" → four (help GCU2BUWJH2W3A8Z7 added, "up to 100% for all placements").
+
+Removals: the unsourced Jay Richman quote; "live image features"; the unsourced "Pacvue company" and "renamed" claims; the unsupported "60 days" default; "no PPC automation" for Jungle Scout; the non-existent `GET /sp/campaigns/{id}/optimizationRules` read.
+
+Pages that stayed unreachable: Pacvue (403); Helium 10 knowledge base (403); Sellozo (JavaScript shell); Skai's Amazon page (image bytes); Intentwise harvesting help article (404); Ad Badger *Bid optimization* (404); m19 pricing page; Perpetua /amazon/ (404); `https://advertising.amazon.com/ads-agent` (sign-in redirect, not followed); the MCP endpoints and the `/skills` download (forbidden hosts).
+
+The verifier called no Ads API host (advertising-api*.amazon.com), no Login with Amazon host, no Amazon Ads MCP endpoint (advertising-ai*.amazon.com) and no SP-API host; it read only public documentation, help, news, vendor, GitHub API, arXiv, amazon.science and Crossref pages by GET, and wrote only this file.
+
+### Gap fill 2026-11-09
+
+Done by the research seat after three completeness critics. Every page named below was opened on 2026-11-09 by plain GET (Amazon Ads news, product, help and expert-advice pages as HTML reduced to text or as embedded help content; Ads API guides as cloudfront markdown; the v1 merged model parsed; aboutamazon and sellingpartners.aboutamazon pages; vendor pages, wire releases and casestudies.com; the AiAds and Holtz–Aral PDFs read in full through PDFKit; Crossref and PubMed records). No Ads API host, Login with Amazon host, MCP endpoint, SP-API host or Seller Central help content call was made.
+
+Added:
+
+1. Gap 1 (Seller Assistant and Project Amelia). New section L: Project Amelia's 2024-09-19 beta; the 2025-09-17 agentic Seller Assistant; Accelerate 2026's drafted SP and SB campaigns and the auto-optimization that adds keywords and product targets, pauses keywords and holds bids "under a maximum the seller sets"; Amazon Ads' own description of the same auto-optimization (one-click switch, reversible actions, US availability January 2027) and the campaign intelligence dashboard (US, October 2026); Seller Assistant workflows and their two descriptions of autonomy; the Claude and Quick plugin; no API field or release note found; how it differs from Ads Agent. Three section A rows, scope lines, Stable 18, run-time reads, open questions 17–20, P-23 to P-26, two seat positions, and pages-read rows. The critic's framing "what it changes without asking" is answered from Amazon's words: auto-optimization's changes are made on the advertiser's behalf and reversible afterwards; drafted campaigns and workflow actions wait for approval; Amazon's pages also say a seller can choose that it "acts on its own", which the file records as a conflict, not a rule.
+2. Gap 2 (Pacvue and Skai). The combined short section is replaced by a Pacvue section (Pacvue Agent, Prism, Dynamic Dayparting from Stream with bids up and down by hour, the snippet-only rules engine and share-of-voice claims, kept as unverified) and a Skai section (Budget Navigator, automated negatives and harvesting, AI dayparting from Stream, automated actions, share of voice, the measurement product still called Impact Navigator in its quotes, holdout guidance, Agent Connect). Pacvue's own pages stayed behind a Cloudflare challenge, so its claims rest on its wire releases and case studies; neither vendor's public pages say what runs by default (open question 21), so "what each automates by default, from their own help pages" could not be filled: both help centres are closed. The one-view table, P-32 and P-33 are added.
+3. Gap 3 (how Amazon's figures were measured). New section M: each figure with its footnote verbatim, marked as a vendor-reported comparison with no holdout described; the rule-based bidding eligibility gate as a selection problem; regression to the mean (Barnett, van der Pols and Dobson 2005); interference (AiAds, full text; Blake and Coey 2014 via Holtz and Aral 2020). Section G's figures bullet now carries the full footnotes ("results are not predictive of future results"). Stable 19–20, P-27 to P-30.
+
+Settled:
+
+4. Contradiction 1 (Ads Agent), owner this file. Re-opened the news page ("the new name of the unified platform"; the 2025 unification of "the demand-side platform and ads console"), the product page (the conversational experience is the closed beta) and help pages GR6VEAG2EQWG56RD and GXKXR56ZCMJ3WCS4 ("Amazon Ads Agent (formerly known as Amazon DSP)" in DSP-specific passages). P-13 now separates the platform name from the closed-beta conversational experience and says the "formerly DSP" wording refers to the same platform; section G gains the help-centre bullet and a seat reading; Stable 11 rewritten. `beyond/ad-products-and-eligibility.md` P-45 is accurate as a quote but narrower than the platform; `areas/budgets-and-pacing.md` P-60 already matches.
+5. Contradiction 2 (Rufus and Alexa+), owner `beyond/ad-products-and-eligibility.md`. Re-opened *Meet Alexa for Shopping* (2026-05-13) and *What agentic shopping means for advertising* (2026-06-11). Rufus was brought together with Alexa+ as Alexa for Shopping on the US app, website and Echo Show; Alexa+ continues on devices and Alexa.com, and Amazon's ads page keeps Alexa+ devices (Alexa+ Conversational Ads, Prime Video Sponsored Tiles) as a separate ad surface. "Alexa for Shopping" does appear on Amazon Ads pages (that page, and the 2026-09-22 small-business page). The owner's current §S already reads this way (Rufus part of Alexa for Shopping in the US, Rufus keeping its name outside the US, Alexa+ on devices), so the critic's "appears on no ads page read" quotes an earlier state of that file. Section F and P-12 aligned, and section F's seat note now scopes the change to the US.
+6. Contradictions 3 and 5 (Reporting v1 concurrency), owner `method/reporting.md`. Re-opened all five pages as markdown. Per advertiser: GA note, *Amazon Ads reporting overview*, and the Stream v2 *Overview*, whose note reads "subject to usage limits (one concurrent report per advertiser, approximately 100 reports per day)". Per client id: *Reporting API v1 overview*, *Limits and best practices*. Section J no longer states one side as the rule; P-31 and Stable 21 added. On the second half of contradiction 5: `method/reporting.md` is right that the Stream v2 overview repeats the per-advertiser line, and `method/change-measurement.md`'s "the Stream v2 overview ... states neither" (its open question 12, its pages-read row "no concurrency statement" and its correction 14) does not match the page as read on 2026-11-09.
+7. Contradictions 4 and 6 (dynamic bidding ceiling), owner `areas/bids-and-placements.md`. Re-opened the library guide ("Currently, we can increase this bid by up to 100% ... for all placements opportunities"), help GCU2BUWJH2W3A8Z7 ("up to 100% for all placements") and the author's guide ("up to 100% for placements at the top of the first page ... up to 50% for all other placements"). The owner's settlement holds: four current Amazon sources say 100% everywhere, one says 100/50, no page dates a change, and "older rule" is a reading. Section B's seat note, Stable 2 and P-2 now name the author's guide as the dissenting Amazon page and defer to P-bids-and-placements-8; the previous seat wording "is the older rule" is withdrawn. `method/change-measurement.md` P-12, `areas/launch.md` P-24 and `areas/diagnosis.md` P-37 as read on 2026-11-09 already name the author's guide's 100/50 beside the 100%, which matches the owner; treating the dating as unresolved is consistent with the owner's settlement.
+
+Corrected while re-opening:
+
+8. Section E: Full-Funnel's 67% and 29% were attributed to *Introducing Amazon Ads Agent*, which does not carry them → sourced to the unBoxed 2026 page and *Grow total sales with Full-Funnel Campaigns*, with the footnote.
+9. Published research: AiAds was cited from its abstract plus a search summary → full text read; the groups were chosen, not randomised; the "abates" line is the authors' expectation, not a measured decline; table 2's revenue (43.51%) differs from the text's "more than 47%".
+
+Declined, in part:
+
+10. Gap 3's "the Baidu AiAds gain shrank as more advertisers adopted it (Blake & Coey 2014)" is wrong on two facts. The AiAds paper (Yang et al. 2019) says the gain "will gradually abate"; it predicts and does not measure a shrink. Blake and Coey (2014) is an eBay email-marketing experiment on test-control interference and does not study Baidu or AiAds. Both are recorded in section M under their own names, and the pairing is not used.
+
+The gap-fill additions above (sections L and M, the Pacvue and Skai sections, P-23 to P-33, Stable 18 to 21, open questions 17 to 21 and the new pages-read rows): Pending verification.