opensquid 0.5.395

package/packs/builtin/seo-aeo-expert/lessons/cloudflare-default-block/lesson.md

@@ -0,0 +1,18 @@
+# Cloudflare blocks AI bots by default — audit before assuming AI visibility
+
+**Confidence: [H]** (Cloudflare primary sources)
+
+Since 2025-07, Cloudflare blocks AI crawlers **by default for new zones**
+(one-click block for existing); pay-per-crawl (HTTP 402) is in closed beta.
+Webflow and other hosts ship similar defaults. Consequence: sites are
+routinely _unintentionally_ invisible to AI engines while their robots.txt
+looks permissive — the CDN eats the bot before robots.txt matters.
+
+**How to apply:** the "am I accidentally blocking citation bots?" audit is
+the single highest-yield AEO check for any Cloudflare/Webflow-fronted site:
+verify AI Crawl Control settings against declared bot policy, then confirm
+with server logs (search-bot user agents actually fetching 200s). Re-check
+after any CDN/WAF change. Bot classes: [[bot-taxonomy-three-classes]].
+
+**Evidence:** https://blog.cloudflare.com/introducing-pay-per-crawl/ ·
+https://developers.cloudflare.com/ai-crawl-control/features/manage-ai-crawlers/

package/packs/builtin/seo-aeo-expert/lessons/deprecated-tactics-2026/lesson.md

@@ -0,0 +1,32 @@
+# Deprecated SEO tactics 2026 — the do-not-encode blocklist
+
+**Confidence: [H]** (each entry traces to a primary source)
+
+1. **llms.txt as a ranking lever** — zero correlation, Google: "purely
+   speculative" ([[llms-txt-dead]]).
+2. **Special AI markup / chunking-for-Google / AI-phrased rewrites** —
+   explicitly unnecessary per Google's official guide
+   ([[google-official-ai-guide]]). Passage structure still helps _other_
+   engines ([[chunk-shape]]) — keep the nuance.
+3. **Backlink volume as the primary AI-visibility KPI** — mentions correlate
+   ~3x stronger ([[mentions-beat-backlinks]]). Demoted, not dead.
+4. **FAQ/HowTo rich-result chasing** — removed 2023; FAQ fully dead 2026-05.
+   Claimed AI-citation lift is vendor folklore.
+5. **Keyword density / meta keywords / LSI / exact-match variation pages** —
+   keyword stuffing measured neutral-to-negative for citations (Princeton
+   GEO); scaled variations hit the scaled-content-abuse policy.
+6. **Pre-2024 CTR curves and rank-only KPIs** — position-1 CTR fell 39.8%
+   (2022) → 27.6% (2026); deep rank tracking unreliable post-num=100
+   ([[measurement-discontinuities]]).
+7. **Featured-snippet ("position zero") programs** — absorbed by AI Overviews.
+8. **Blanket-blocking all AI bots "to protect content"** — kills citations;
+   distinguish bot classes ([[bot-taxonomy-three-classes]]).
+9. **Reddit as a guaranteed citation hack** — mix is volatile; spam-seeding
+   filtered ([[ugc-citation-channels]]).
+10. **"AEO is a separate discipline with a separate stack"** — Google frames
+    it as SEO; the delta is real (mentions, chunks, Bing, UGC) but the
+    foundation is shared.
+
+**How to apply:** any recommendation matching this list needs either a
+primary-source rebuttal or removal. The stale-tactic-guard skill enforces
+the regex-detectable subset at write time.

package/packs/builtin/seo-aeo-expert/lessons/engine-divergence/lesson.md

@@ -0,0 +1,22 @@
+# AI engines barely overlap — route per engine
+
+**Confidence: [M]** (680M-citation analysis; vendor methodology)
+
+Only ~**11% of cited domains overlap between ChatGPT and Perplexity**. Each
+engine is its own ecosystem:
+
+| Engine               | Retrieval backbone          | Distinct lever                             |
+| -------------------- | --------------------------- | ------------------------------------------ |
+| Google AI Mode / AIO | Google index + core ranking | entity authority, info gain, GBP           |
+| ChatGPT Search       | Bing index + OAI-SearchBot  | Bing Webmaster Tools ([[bing-is-chatgpt]]) |
+| Perplexity           | own index, Reddit/UGC-heavy | Yelp, niche directories, Reddit presence   |
+| Gemini               | Google + GBP local data     | profile accuracy (best local recs)         |
+| Claude               | Brave + Claude-SearchBot    | structured deep content                    |
+
+**How to apply:** a single "AI visibility" number is meaningless — sample
+citation share per engine on priority queries, and invest per-engine where
+the audience is. One Google parameter change once dropped ChatGPT's Reddit
+share 60%→10% in six weeks — the mix is volatile; re-sample, don't assume.
+
+**Evidence:** https://authoritytech.io/curated/ai-citation-11-percent-platform-overlap-per-engine-audit-2026 ·
+https://discoveredlabs.com/blog/chatgpt-claude-perplexity-and-google-ai-overviews-how-each-platform-cites-sources-differently

package/packs/builtin/seo-aeo-expert/lessons/freehand-schema-forbidden/lesson.md

@@ -0,0 +1,21 @@
+# Freehand LLM-generated JSON-LD is forbidden — validator before ship
+
+**Confidence: [M]** (practitioner consensus, consistent across 2026 reviews)
+
+The 2026 agent-SEO consensus: "AI is absolutely horrible at writing schema"
+freehand — hallucinated properties, wrong enum values, missing required
+fields (priceValidUntil, reviewCount), invalid nesting. The failures are
+silent: pages render fine while rich-result eligibility quietly dies.
+
+**How to apply (the rule):** every JSON-LD change passes a validator before
+push — `ajv` against schema.org definitions, Google Rich Results Test, or
+Schema.org validator. Product offers additionally need priceValidUntil +
+MerchantReturnPolicy + OfferShippingDetails for Merchant eligibility.
+AggregateRating needs reviewCount/ratingCount. Schema must be in
+server-rendered HTML ([[no-js-execution]]). The schema-validation-gate skill
+
+- ops FSM enforce this: JSON-LD edits open the cycle, a validator run closes
+  it, pushing while open gets flagged.
+
+**Evidence:** https://www.frase.io/blog/ai-agents-for-seo ·
+https://aimultiple.com/seo-ai (agent-workflow reliability reviews, 2026)

package/packs/builtin/seo-aeo-expert/lessons/geo-content-levers/lesson.md

@@ -0,0 +1,20 @@
+# GEO content levers: statistics, expert quotes, cited sources
+
+**Confidence: [M]** (Princeton GEO, KDD 2024 — direction replicated on
+production engines by multiple 2025-26 practitioner studies; exact
+percentages are benchmark-conditional)
+
+Of 9 interventions tested, the winners for AI citation/visibility:
+**adding statistics (~+40%)**, **direct quotes from named sources (~+28%)**,
+**citing authoritative sources (+30-40%)**. Keyword stuffing measured
+neutral-to-**negative**. Fluency improvements also helped (~+29%).
+
+**How to apply:** every content brief should require: concrete numbers with
+sources, at least one named-expert quote, outbound citations to primary
+sources, and extractable facts (dates, named entities). This composes with
+[[chunk-shape]] — the stats/quotes need to live inside self-contained
+sections to be retrieved.
+
+**Evidence:** Aggarwal et al., KDD 2024 (Princeton GEO) ·
+https://seenrank.com/blog/the-princeton-geo-study-explained-for-marketers/ ·
+https://sunilpratapsingh.com/guides/geo/what-research-says-about-generative-engine-optimization

package/packs/builtin/seo-aeo-expert/lessons/google-official-ai-guide/lesson.md

@@ -0,0 +1,17 @@
+# Google's official AI guide: AEO is still SEO — and what Google says NOT to do
+
+**Confidence: [H]** (Google primary source, 2026-05-15)
+
+Google's first official "optimizing for generative AI in Search" guide states
+AI features are "rooted in our core Search ranking and quality systems" —
+AEO/GEO is framed as SEO. Explicitly **not needed for Google**: llms.txt /
+"AI text files", special AI markup, breaking content into chunks, rewriting
+in "AI phrasing", inauthentic mention-building. **Recommended:** first-hand
+non-commodity content, crawlable technically-sound pages, semantic
+human-readable HTML, Merchant Center feeds, Google Business Profile.
+
+**Caveat to keep:** the "don't chunk" claim is about _Google's_ systems.
+Passage-level structure still demonstrably helps ChatGPT/Perplexity/Claude
+retrieval ([[chunk-shape]]). Encode both views, don't flatten the nuance.
+
+**Evidence:** https://developers.google.com/search/docs/fundamentals/ai-optimization-guide (2026-05-15)

package/packs/builtin/seo-aeo-expert/lessons/llms-txt-dead/lesson.md

@@ -0,0 +1,19 @@
+# llms.txt is dead as a ranking/citation lever
+
+**Confidence: [H]** on non-adoption (multiple independent confirmations)
+
+No major AI platform reads llms.txt as a first-class retrieval input. The
+record: Google's official guide says it's not needed (John Mueller: "purely
+speculative"); SE Ranking's 300k-domain study found **zero correlation**
+with AI citations (removing the feature _improved_ their model's accuracy);
+one analysis found 408 of 500M AI-bot visits touched the file; OpenAI makes
+no commitment to read it. Site adoption plateaued at ~5-15%.
+
+**How to apply:** never recommend llms.txt for ranking/citation purposes.
+It remains harmless and mildly useful for _agent workflows_ (IDE agents,
+docs-for-LLMs navigation) — that's the only honest pitch. Time freed goes to
+levers with evidence: [[geo-content-levers]], [[mentions-beat-backlinks]].
+
+**Evidence:** https://www.searchenginejournal.com/google-says-llms-txt-is-purely-speculative-for-now/577576/ ·
+https://theseocommunity.com/resources/blog/llms-txt-should-we-or-not ·
+https://developers.google.com/search/docs/fundamentals/ai-optimization-guide

package/packs/builtin/seo-aeo-expert/lessons/local-ai-gap/lesson.md

@@ -0,0 +1,20 @@
+# The local AI gap: ChatGPT recommends 1.2% of local businesses
+
+**Confidence: [M-H]** (SOCi 2026 Local Visibility Index + Google primary docs)
+
+ChatGPT recommends only **1.2%** of local businesses (vs Google's 35.9%).
+Driver: profile-data accuracy is ~68% on ChatGPT/Perplexity vs ~100% on
+Gemini — which is why Gemini's local recommendation rate is ~10x higher.
+AI Overview local packs are replacing the classic 3-pack on ~8% of local
+queries (climbing), often surfacing only 1-2 businesses.
+
+**The fix is mechanical (profile stack):** Google Business Profile (Google's
+official guide names it a structured AI input) + **Bing Places** (feeds
+ChatGPT) + Apple Maps + Yelp (feeds Perplexity) + consistent NAP everywhere +
+review volume/recency + service pages answering specific local questions +
+LocalBusiness schema. This is the most underserved, most mechanizable AEO
+segment — commercial tools price it out at $250+/mo.
+
+**Evidence:** https://www.soci.ai/blog/how-to-rank-in-chatgpt-perplexity-and-google-ai-overview/ ·
+https://www.searchable.com/blog/local-business-ai-search-optimization-2026 ·
+https://www.sterlingsky.ca/the-state-of-local-seo-in-2026/

package/packs/builtin/seo-aeo-expert/lessons/measurement-discontinuities/lesson.md

@@ -0,0 +1,29 @@
+# Measurement discontinuities and the 2026 tracking stack
+
+**Confidence: [H]** (Google primary sources + replicated analyses)
+
+Two hard discontinuities make naive trend analysis wrong:
+
+1. **2025-09 `num=100` removal** — 87.7% of sites lost GSC impressions and
+   average position "improved" overnight (bot impressions removed). Artifact,
+   not ranking change.
+2. **2026-05 AI Mode default** — click/CTR baselines shift again.
+
+GSC's **"Generative AI performance" report** (2026-06, data from 2026-05-18):
+impressions only — no clicks, no queries, no API/BigQuery; manual CSV export.
+Before it, AI Mode data was silently lumped into "Web" totals.
+
+**LLM referrals:** typically <2% of referral traffic but convert ~3x
+(Microsoft Clarity; multiple datasets). Track in GA4 with referrer regex
+(chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com) —
+note Gemini/AI-Mode referrals are largely indistinguishable from google.com
+organic. Referral mix is shifting: ChatGPT 86.7%→64.5% of LLM referrals in a
+year; Gemini 5.7%→21.5%.
+
+**How to apply:** annotate both discontinuity dates in every dashboard;
+never compare across them without correction; add the GA4 LLM-referral
+segment before any AEO work so before/after is measurable.
+
+**Evidence:** https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports ·
+https://zeo.org/resources/blog/the-impact-of-num-100-parameter-removal-on-seo-reporting ·
+https://clarity.microsoft.com/blog/ai-traffic-converts-at-3x-the-rate-of-other-channels-study/

package/packs/builtin/seo-aeo-expert/lessons/mentions-beat-backlinks/lesson.md

@@ -0,0 +1,16 @@
+# Unlinked brand mentions out-correlate backlinks ~3x for AI visibility
+
+**Confidence: [H]** (Ahrefs 75,000-brand study)
+
+Spearman correlation with AI citation presence (AI Overviews, ChatGPT, AI
+Mode): **unlinked web mentions 0.664** · YouTube mentions **0.737** (the
+strongest single predictor) · brand search volume 0.334 · **backlinks 0.218**.
+The "mention-source divide": AI often recommends _your brand_ while citing a
+_third-party page_ (G2, Reddit, press, directories).
+
+**How to apply:** earned third-party presence (reviews, listicles, press,
+YouTube, communities) is a first-class AEO tactic, co-equal with on-site
+work. Backlinks are demoted, not dead — they still feed classic ranking,
+which feeds RAG retrieval. Budget accordingly.
+
+**Evidence:** https://ahrefs.com/blog/ai-overview-brand-correlation/ (2025-2026)

package/packs/builtin/seo-aeo-expert/lessons/no-js-execution/lesson.md

@@ -0,0 +1,18 @@
+# No major AI crawler executes JavaScript — SSR/SSG is a hard gate
+
+**Confidence: [H]** (replicated; Vercel study + 500M-fetch analysis)
+
+Analysis of 500M+ GPTBot fetches found **zero JS execution**; Vercel's AI
+crawler study (2024-12) established it and it remains true in 2026.
+Googlebot is the only full-rendering crawler; Bingbot renders partially —
+and ChatGPT depends on Bing's index. Client-side-rendered content is
+therefore invisible to ChatGPT, Claude, and Perplexity.
+
+**How to apply:** indexable content must be in the server-rendered HTML
+(SSR/SSG). In Next.js: pages and layouts stay server components; `"use
+client"` belongs at the interactive leaf. Verify with `curl` (no JS) — what
+you see is what AI engines see. JSON-LD must also be server-side
+([[schema-actual-role]]).
+
+**Evidence:** https://vercel.com/blog/the-rise-of-the-ai-crawler ·
+https://www.asklantern.com/blogs/ai-crawlers-do-not-render-javascript

package/packs/builtin/seo-aeo-expert/lessons/schema-actual-role/lesson.md

@@ -0,0 +1,22 @@
+# Schema's actual role in AI search — useful, but not where vendors claim
+
+**Confidence: [H]** on the confirmed parts; **[C]** flagged on the folklore
+
+**Confirmed:** Microsoft's Fabrice Canel (SMX Munich 2025-03): Bing/Copilot
+**do use schema markup to help their LLMs understand content**. Schema
+remains required for rich results and Merchant Center eligibility, and aids
+entity disambiguation. **Also confirmed:** Google's official guide —
+structured data is **not required** for Google AI features.
+
+**Folklore (do not encode):** "schema = 2.5x citation chance", "40% more
+AIO appearances", "schema is the language of LLM tokenization" — no credible
+methodology behind any of these (the tokenization claim is technically false).
+
+**How to apply:** ship schema for rich results, Merchant Center, Bing/Copilot
+comprehension, and entity grounding — server-side rendered
+([[no-js-execution]]), validator-checked ([[freehand-schema-forbidden]]).
+Never sell it internally as a Google AI-citation lever.
+
+**Evidence:** https://searchengineland.com/microsoft-bing-copilot-use-schema-for-its-llms-453455 ·
+https://developers.google.com/search/docs/fundamentals/ai-optimization-guide ·
+https://searchengineland.com/schema-markup-ai-search-no-hype-472339

package/packs/builtin/seo-aeo-expert/lessons/schema-claims-need-ground-truth/lesson.md

@@ -0,0 +1,34 @@
+# Schema is claim-bearing — syntactic validation cannot catch fiction
+
+**Confidence: [H]** (first-principles + live incident, 2026-06-10)
+
+Structured data sits at the intersection where "technical SEO" stops being
+business-agnostic: every Offer, Service, price, availability, and
+description in JSON-LD is a **statement of business fact**. Validators
+(ajv, Rich Results Test, Schema.org) check the _syntactic_ layer only.
+Among claim types, only dates self-falsify (`validThrough` in the past).
+Wrong prices, nonexistent products, and wrong capacities pass every
+validator — and AI engines extract and repeat them verbatim.
+
+**The incident that proved it:** a boutique-studio site served a
+syntactically perfect OfferCatalog advertising a product that did not exist
+("Group — $350 one-month unlimited") with wrong rates across the board —
+generated from a stale code-side data table, invisible on the rendered page
+(client-rendered UI), visible to every AI engine, for weeks. No validator
+or HTML audit could have caught it; only the owner could.
+
+**How to apply — the claims-free / claim-bearing boundary:**
+
+- Claims-free (canonicals, hreflang, title length, redirects, rendering):
+  verify mechanically, ship on green checks.
+- Claim-bearing (any schema node describing offerings/prices/people/hours):
+  every value must trace to a **ground-truth source** — the live
+  API/database the visible UI renders from, or explicit owner confirmation.
+  Never a parallel static file ([[freehand-schema-forbidden]] covers shape;
+  this lesson covers TRUTH).
+- Before auditing a site's offerings, find where the catalog actually
+  lives. If the UI is client-rendered, the HTML lies by omission — to
+  crawlers AND to you ([[no-js-execution]]).
+- Maintain a business ground-truth doc (real lineup, hard rules, known
+  anti-sources) loaded at session start; schema work without it is
+  fiction-risk, not SEO.

package/packs/builtin/seo-aeo-expert/lessons/ugc-citation-channels/lesson.md

@@ -0,0 +1,22 @@
+# UGC citation channels: Reddit, YouTube, Wikipedia — powerful and volatile
+
+**Confidence: [M]** (large citation analyses; mix shifts fast)
+
+The citation economics as of early 2026: **YouTube overtook Reddit** as the
+#1 social source in LLM answers (16% vs 10% of answers — a reversal from
+2025); Google AI Overviews still cite Reddit in ~21% of responses; Perplexity
+remains Reddit-heavy (one study: 46.7% of its top citations);
+Wikipedia+Reddit exceed 25% of US ChatGPT citations. Reddit citation share
+grew 73% in commercial categories (Tinuiti Q1 2026). Volatility warning: one
+Google parameter change dropped ChatGPT's Reddit share 60%→10% in six weeks.
+
+**How to apply:** legitimate Reddit presence and transcribed, demonstrative
+YouTube video are _citation channels_, not just social channels — YouTube
+mentions are the single strongest AI-visibility correlate (0.737,
+[[mentions-beat-backlinks]]). Spammy Reddit seeding is being filtered;
+authentic participation only. Re-sample the mix quarterly
+([[engine-divergence]]).
+
+**Evidence:** https://searchengineland.com/ai-search-engines-cite-reddit-youtube-and-linkedin-most-study-473138 ·
+https://wellows.com/blog/social-media-ai-citations-report-2026/ ·
+https://contently.com/2026/04/29/top-sources-llms-cite/

package/packs/builtin/seo-aeo-expert/manifest.yaml

@@ -0,0 +1,128 @@
+# seo-aeo-expert — evidence-gated SEO/AEO discipline as a domain pack.
+# Spec: ~/projects/loop/docs/opensquid-seo-aeo-pack-spec.md (2026-06-09).
+# Thesis: every commercial AEO tool observes; knowledge skills advise; this pack
+# ENFORCES at write-time (guards), remembers (seed lessons), and disciplines the
+# audit → fix → validate → ship → measure loop (fsm.yaml). Deep how-to lives in
+# the seo-* Claude skill suite — the pack never duplicates it.
+name: seo-aeo-expert
+version: 0.2.0
+scope: domain
+goal: >-
+  Enforce evidence-based SEO/AEO discipline while building and operating
+  websites — block visibility-destroying changes (citation-bot disallows,
+  robots wildcards), warn on 2026-deprecated tactics and AI-invisible
+  rendering, and gate ship-time on validation evidence via the SEO ops FSM.
+description: >-
+  Domain pack for SEO + AEO/GEO (AI search) work, June-2026 evidence base.
+  Three runtime layers: 18 seed lessons (knowledge, each with primary-source
+  URL + confidence tag), deterministic guards (discipline), and the SEO ops
+  lifecycle FSM (audit/fix/validate/ship/measure). Blocks are reserved for
+  irreversible invisible damage (robots-level de-indexing, citation-bot
+  blocking); everything else warns — SEO is judgment-heavy, so the pack
+  informs and the human decides. Never encodes vendor folklore: every rule
+  traces to a primary source recorded in its lesson.
+evolves: true
+foundation:
+  domains:
+    - seo
+    - aeo-geo
+    - web-marketing
+    - content
+  methodologies:
+    - evidence-gated-recommendations
+    - answer-first-content
+    - entity-seo
+
+# No detected_by gate: opt-in via active.json IS the pin (same decision as
+# coding-flow, for the same EWG.3.1 reason — found live 2026-06-10). The IDF.3
+# DetectionContext only pre-stages 5 well-known root files (package.json,
+# tsconfig.json, Cargo.toml, pyproject.toml, go.mod) and leaves recentPrompts
+# empty, so file_glob clauses like `*/next.config.*` and conversation_signal
+# clauses can NEVER match today — any such clause silently DISABLES an
+# opted-in pack (verified: dispatch loaded packs=4 without this pack until
+# the clauses were removed). Off-domain noise is acceptable because every
+# rule already self-gates on file paths (robots.txt, page.tsx, JSON-LD
+# content) before any verdict. Revisit when IDF.3 Phase 2/3 stages globs +
+# prompt history.
+
+# L1 — the knowledge layer. Each lesson: claim → evidence (URL + date) →
+# confidence ([H]/[M]/[C]) → how to apply. Idempotent ingest (external_id =
+# sha256(pack@version|title)); this list is the pack's upgrade surface — when
+# evidence changes, edit the lesson + bump version.
+seed_lessons:
+  - title: 'AI Mode is the default Google experience — optimize for query fan-out'
+    body_path: lessons/ai-mode-default-fanout/lesson.md
+    tags: [seo, aeo, google, ai-mode]
+    source: 'https://developers.google.com/search/docs/fundamentals/ai-optimization-guide'
+  - title: 'Google official AI guide: AEO is still SEO; llms.txt/AI-markup/chunking not needed for Google'
+    body_path: lessons/google-official-ai-guide/lesson.md
+    tags: [seo, aeo, google]
+    source: 'https://developers.google.com/search/docs/fundamentals/ai-optimization-guide'
+  - title: 'Get cited, not just ranked — citation inside AI answers recovers clicks'
+    body_path: lessons/citation-over-rank/lesson.md
+    tags: [seo, aeo, ctr]
+    source: 'https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/'
+  - title: 'Unlinked brand mentions out-correlate backlinks ~3x for AI visibility'
+    body_path: lessons/mentions-beat-backlinks/lesson.md
+    tags: [seo, aeo, links, mentions]
+    source: 'https://ahrefs.com/blog/ai-overview-brand-correlation/'
+  - title: 'Three AI bot classes: training vs search/citation vs user-fetch — per-class policy'
+    body_path: lessons/bot-taxonomy-three-classes/lesson.md
+    tags: [seo, aeo, bots, robots-txt]
+    source: 'https://developers.openai.com/api/docs/bots'
+  - title: 'Cloudflare blocks AI bots by default since 2025-07 — audit before assuming AI visibility'
+    body_path: lessons/cloudflare-default-block/lesson.md
+    tags: [seo, aeo, bots, cloudflare]
+    source: 'https://blog.cloudflare.com/introducing-pay-per-crawl/'
+  - title: 'No major AI crawler executes JavaScript — SSR/SSG is a hard gate for AI visibility'
+    body_path: lessons/no-js-execution/lesson.md
+    tags: [seo, aeo, rendering, ssr]
+    source: 'https://vercel.com/blog/the-rise-of-the-ai-crawler'
+  - title: 'ChatGPT Search is the Bing index — Bing Webmaster Tools is a prerequisite'
+    body_path: lessons/bing-is-chatgpt/lesson.md
+    tags: [seo, aeo, bing, chatgpt]
+    source: 'https://www.soci.ai/blog/how-to-rank-in-chatgpt-perplexity-and-google-ai-overview/'
+  - title: 'AI engines barely overlap (~11% ChatGPT/Perplexity) — route per engine'
+    body_path: lessons/engine-divergence/lesson.md
+    tags: [seo, aeo, engines]
+    source: 'https://authoritytech.io/curated/ai-citation-11-percent-platform-overlap-per-engine-audit-2026'
+  - title: 'GEO content levers: statistics, expert quotes, cited sources (Princeton GEO)'
+    body_path: lessons/geo-content-levers/lesson.md
+    tags: [seo, aeo, content, geo]
+    source: 'https://seenrank.com/blog/the-princeton-geo-study-explained-for-marketers/'
+  - title: 'Chunk shape: 200-400-word self-contained sections, answer-first openings'
+    body_path: lessons/chunk-shape/lesson.md
+    tags: [seo, aeo, content]
+    source: 'https://www.lumar.io/blog/best-practice/content-chunking-ai-extractability-geo-aeo-explainer/'
+  - title: 'Schema actual role: Bing/Copilot comprehension + rich results yes; Google AI-citation lever no'
+    body_path: lessons/schema-actual-role/lesson.md
+    tags: [seo, aeo, schema, structured-data]
+    source: 'https://searchengineland.com/microsoft-bing-copilot-use-schema-for-its-llms-453455'
+  - title: 'llms.txt is dead as a ranking lever — zero correlation across 300k domains'
+    body_path: lessons/llms-txt-dead/lesson.md
+    tags: [seo, aeo, llms-txt]
+    source: 'https://www.searchenginejournal.com/google-says-llms-txt-is-purely-speculative-for-now/577576/'
+  - title: 'UGC citation channels: Reddit/YouTube/Wikipedia economics, volatile mix'
+    body_path: lessons/ugc-citation-channels/lesson.md
+    tags: [seo, aeo, ugc, reddit, youtube]
+    source: 'https://searchengineland.com/ai-search-engines-cite-reddit-youtube-and-linkedin-most-study-473138'
+  - title: 'Local AI gap: ChatGPT recommends 1.2% of local businesses — profile stack is the fix'
+    body_path: lessons/local-ai-gap/lesson.md
+    tags: [seo, aeo, local, gbp]
+    source: 'https://www.soci.ai/blog/how-to-rank-in-chatgpt-perplexity-and-google-ai-overview/'
+  - title: 'Measurement discontinuities: Sept-2025 num=100, May-2026 AI-Mode default, GSC GenAI limits'
+    body_path: lessons/measurement-discontinuities/lesson.md
+    tags: [seo, aeo, analytics, gsc, ga4]
+    source: 'https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports'
+  - title: 'Deprecated SEO tactics 2026 — the do-not-encode blocklist'
+    body_path: lessons/deprecated-tactics-2026/lesson.md
+    tags: [seo, aeo, deprecated]
+    source: 'https://developers.google.com/search/docs/fundamentals/ai-optimization-guide'
+  - title: 'Freehand LLM-generated JSON-LD is forbidden — validator before ship'
+    body_path: lessons/freehand-schema-forbidden/lesson.md
+    tags: [seo, schema, validation]
+    source: 'https://www.frase.io/blog/ai-agents-for-seo'
+  - title: 'Schema is claim-bearing — syntactic validation cannot catch fiction; every value needs a ground-truth source'
+    body_path: lessons/schema-claims-need-ground-truth/lesson.md
+    tags: [seo, schema, ground-truth, aeo]
+    source: 'live incident, raumpilates 2026-06-10 — stale OfferCatalog served fictional product for weeks'

package/packs/builtin/seo-aeo-expert/models.yaml

@@ -0,0 +1,19 @@
+# seo-aeo-expert model aliases — alias names + call modes ONLY, never literal
+# model names (user binds alias → concrete model in ~/.opensquid/models.yaml).
+# P0 ships deterministic guards only; these aliases are declared now so the
+# P1 skills (answer-first-content-check, evidence-gated-recs, destination
+# check) land without a manifest change.
+fast_classifier:
+  description: >-
+    Single-label classification: folklore-vs-evidenced recommendation check,
+    pack destination check. Suggested tier: fast.
+  mode: subscription
+  impl: cli
+
+content_judge:
+  description: >-
+    Short structured judgement on content shape (answer-first opening,
+    extractable facts) and validation-evidence audits via cached_audit.
+    Suggested tier: fast-to-mid.
+  mode: subscription
+  impl: cli

package/packs/builtin/seo-aeo-expert/skills/bot-policy-guard/skill.yaml

@@ -0,0 +1,159 @@
+# bot-policy-guard — the highest-value guard. Cloudflare/Webflow default-block
+# (2025-07) makes ACCIDENTAL AI invisibility the #1 silent AEO failure. Three
+# rules, by damage class:
+#   citation-bot-disallow   BLOCK  blocking a search/citation bot removes the
+#                                  site from that engine's answers entirely
+#   robots-wildcard-disallow BLOCK Disallow: / under User-agent: * = de-index
+#   training-bot-change     warn   blocking training bots is a legitimate
+#                                  choice — surface the trade-off, don't block
+# Evidence: lessons [[bot-taxonomy-three-classes]] + [[cloudflare-default-block]].
+name: bot-policy-guard
+load: preload
+triggers:
+  - kind: tool_call
+rules:
+  - id: citation-bot-disallow
+    process:
+      - call: text_pattern_match
+        as: is_write
+        args:
+          text_field: tool
+          patterns:
+            - '^(Write|Edit|NotebookEdit)$'
+      - call: text_pattern_match
+        as: botfile
+        if: 'is_write.matched.length > 0'
+        args:
+          text_field: args.file_path
+          patterns:
+            - 'robots\.txt$'
+            - 'vercel\.json$'
+            - 'middleware\.(ts|js)$'
+            - '\.htaccess$'
+            - 'nginx[^/]*\.conf$'
+            - '_headers$'
+            - 'wrangler\.toml$'
+      - call: text_pattern_match
+        as: write_hit
+        if: 'botfile.matched.length > 0'
+        args:
+          text_field: args.content
+          patterns:
+            - 'user-agent:\s*(OAI-SearchBot|Claude-SearchBot|PerplexityBot|Bingbot|Googlebot)\b(?:(?!user-agent)[\s\S]){0,200}?disallow:\s*/'
+            - '(deny|block)[^\n]{0,120}\b(OAI-SearchBot|Claude-SearchBot|PerplexityBot|Bingbot|Googlebot)\b'
+            - '\b(OAI-SearchBot|Claude-SearchBot|PerplexityBot|Bingbot|Googlebot)\b[^\n]{0,120}(deny|block)'
+      - call: text_pattern_match
+        as: edit_hit
+        if: 'botfile.matched.length > 0'
+        args:
+          text_field: args.new_string
+          patterns:
+            - 'user-agent:\s*(OAI-SearchBot|Claude-SearchBot|PerplexityBot|Bingbot|Googlebot)\b(?:(?!user-agent)[\s\S]){0,200}?disallow:\s*/'
+            - '(deny|block)[^\n]{0,120}\b(OAI-SearchBot|Claude-SearchBot|PerplexityBot|Bingbot|Googlebot)\b'
+            - '\b(OAI-SearchBot|Claude-SearchBot|PerplexityBot|Bingbot|Googlebot)\b[^\n]{0,120}(deny|block)'
+      - call: verdict
+        if: 'botfile.matched.length > 0 && (write_hit.matched.length > 0 || edit_hit.matched.length > 0)'
+        args:
+          level: block
+          message: >-
+            seo-aeo-expert/bot-policy: this blocks an AI SEARCH/CITATION bot
+            (OAI-SearchBot, Claude-SearchBot, PerplexityBot, Bingbot, or
+            Googlebot) — that removes the site from the corresponding engine's
+            answers entirely. Distinct from TRAINING bots (GPTBot, ClaudeBot,
+            Google-Extended), which are fine to block. ChatGPT visibility also
+            dies with Bingbot (its index is Bing). See lesson
+            [[bot-taxonomy-three-classes]]. If genuinely intentional, the user
+            must override.
+
+  - id: robots-wildcard-disallow
+    process:
+      - call: text_pattern_match
+        as: is_write
+        args:
+          text_field: tool
+          patterns:
+            - '^(Write|Edit|NotebookEdit)$'
+      - call: text_pattern_match
+        as: robotsfile
+        if: 'is_write.matched.length > 0'
+        args:
+          text_field: args.file_path
+          patterns:
+            - 'robots\.txt$'
+      - call: text_pattern_match
+        as: wild_w
+        if: 'robotsfile.matched.length > 0'
+        args:
+          text_field: args.content
+          patterns:
+            - 'user-agent:\s*\*(?:(?!user-agent)[\s\S]){0,300}?\ndisallow:\s*/[ \t]*(\n|$)'
+      - call: text_pattern_match
+        as: wild_e
+        if: 'robotsfile.matched.length > 0'
+        args:
+          text_field: args.new_string
+          patterns:
+            - 'user-agent:\s*\*(?:(?!user-agent)[\s\S]){0,300}?\ndisallow:\s*/[ \t]*(\n|$)'
+      - call: verdict
+        if: 'robotsfile.matched.length > 0 && (wild_w.matched.length > 0 || wild_e.matched.length > 0)'
+        args:
+          level: block
+          message: >-
+            seo-aeo-expert/bot-policy: `Disallow: /` under `User-agent: *`
+            de-indexes the ENTIRE site from every crawler — Google, Bing, and
+            all AI engines. This is full search invisibility. If this is a
+            staging host, scope it by hostname/environment instead. User must
+            override to proceed.
+
+  - id: training-bot-change
+    process:
+      - call: text_pattern_match
+        as: is_write
+        args:
+          text_field: tool
+          patterns:
+            - '^(Write|Edit|NotebookEdit)$'
+      - call: text_pattern_match
+        as: botfile
+        if: 'is_write.matched.length > 0'
+        args:
+          text_field: args.file_path
+          patterns:
+            - 'robots\.txt$'
+            - 'vercel\.json$'
+            - 'middleware\.(ts|js)$'
+            - '\.htaccess$'
+            - 'nginx[^/]*\.conf$'
+            - '_headers$'
+            - 'wrangler\.toml$'
+      - call: text_pattern_match
+        as: train_w
+        if: 'botfile.matched.length > 0'
+        args:
+          text_field: args.content
+          patterns:
+            - '\b(GPTBot|ClaudeBot|Google-Extended|Bytespider|Meta-ExternalAgent|CCBot|anthropic-ai)\b'
+      - call: text_pattern_match
+        as: train_e
+        if: 'botfile.matched.length > 0'
+        args:
+          text_field: args.new_string
+          patterns:
+            - '\b(GPTBot|ClaudeBot|Google-Extended|Bytespider|Meta-ExternalAgent|CCBot|anthropic-ai)\b'
+      - call: verdict
+        if: 'botfile.matched.length > 0 && (train_w.matched.length > 0 || train_e.matched.length > 0)'
+        args:
+          level: warn
+          message: >-
+            seo-aeo-expert/bot-policy: TRAINING-bot policy touched (GPTBot /
+            ClaudeBot / Google-Extended / Bytespider / Meta-ExternalAgent /
+            CCBot). Blocking these is a legitimate content-protection choice
+            and does NOT affect search/citation visibility — Google-Extended
+            does not control AI Overviews (Googlebot does), and GPTBot does
+            not control ChatGPT Search (OAI-SearchBot does). Know which class
+            you are changing: [[bot-taxonomy-three-classes]].
+prose: |
+  Bot-policy guard. Citation-bot disallows and robots wildcards are blocked
+  because the damage is invisible until traffic is already gone (and AI
+  engines re-crawl slowly — recovery is weeks). Training-bot changes warn
+  with the class distinction so the trade-off is a decision, not an accident.