@damiangilgonzalez/harness 1.3.7

package/skills/seo-geo/SKILL.md

@@ -0,0 +1,192 @@
+---
+name: seo-geo
+description: "Use when one existing page needs to rank in Google AND get cited by AI answer engines — auditing a URL for on-page SEO, structured-data JSON-LD, GEO citation levers, Core Web Vitals, crawlability, and AI-bot access in robots.txt. NOT writing the article body (that is `article-writing`), NOT keyword research or content strategy (that is `content-engine`)."
+tags: [seo, geo, structured-data, schema-org, core-web-vitals, ai-search, technical-seo]
+recommends: [content-engine, article-writing, landing-copy, performance, nextjs, accessibility, analytics]
+origin: risco
+---
+
+# seo-geo — make one page findable in Google AND in AI answers
+
+> One audited page, two surfaces. You take an **existing** URL and make it rank in
+> Google's blue links / AI Overviews **and** earn citations in ChatGPT Search,
+> Perplexity, and Gemini — through meta, schema, answer-first structure, and a clean
+> technical audit. You do **not** write the article body and you do **not** pick the
+> keywords. That work belongs upstream.
+
+## When to use / When NOT to use
+
+**Use when** the request is about a concrete artifact:
+
+- "Audit this page for SEO" / "why isn't this ranking?" / "make this page rank."
+- "Get us cited in ChatGPT / Perplexity / AI Overviews" / "optimize for AI search."
+- "Add schema markup" / "write the JSON-LD for this article/product."
+- "Our Core Web Vitals are failing" / crawlability / "should we block GPTBot?"
+
+**Do NOT use — redirect instead:**
+
+| Request | Owner |
+|---|---|
+| Write the long-form article/blog body | `article-writing` |
+| Keyword research, topic clusters, content calendar | `content-engine` |
+| Write persuasive value-prop / CTA copy | `landing-copy` |
+| Actually fix a slow LCP/INP (render path, hydration, bundle) | `performance` |
+| Wire Next.js `generateMetadata` / app-router SEO plumbing | [`../nextjs/SKILL.md`](../nextjs/SKILL.md) |
+| WAI-ARIA / keyboard / contrast audit | `accessibility` |
+| Measure traffic / set up GA4 / dashboards after the fact | `analytics` |
+
+The load-bearing distinction: **`content-engine`** decides *what to write and which
+keywords to target* (upstream strategy); you optimize the *specific page that exists*.
+**`performance`** owns *why* a render is slow; you only *measure* Core Web Vitals against
+the p75 thresholds, flag a fail, and hand off the fix.
+
+## The one rule both surfaces share
+
+**Answer the primary query completely in the first ~40–200 words.** Everything below is
+downstream of this. Why: Google extracts that opening for featured snippets and AI
+Overviews, and the Princeton GEO study (KDD 2024, 10k queries across 8 domains) found
+answer-first, source-cited content is what AI engines lift into their answers — not
+buried prose. If the page makes a reader scroll past an intro to reach the answer, no
+amount of schema saves it.
+
+There is **no separate "AI SEO" toggle for Google.** AI Overviews and AI Mode run on the
+*same* Googlebot crawl, the same rendering, the same ranking, and the same E-E-A-T /
+helpful-content signals (folded into core ranking in March 2024). Good page experience
+and distinguishable main content serve both surfaces at once. Treat them as one job.
+
+## 1. On-page SEO pass
+
+- **Title ≤ 60 chars**, primary query near the front. Past ~60 chars Google truncates it.
+- **Meta description ≤ 160 chars** that earns the click — it is not a ranking factor, it
+  is the ad copy for the result.
+- **Exactly one `<h1>`**, then a logical `<h2>`/`<h3>` hierarchy that mirrors search
+  intent (the sub-questions a searcher actually has).
+- **Internal links with descriptive anchors** ("Core Web Vitals thresholds", not "click
+  here") — anchor text is an entity signal both Google and AI engines read.
+
+```html
+<!-- Bad: vague, query buried, no length discipline -->
+<title>Home | Our Awesome Company - Welcome to the Best Site Ever for You</title>
+<meta name="description" content="We are a company. Learn more about us here today.">
+
+<!-- Good: query first, within limits, earns the click -->
+<title>Core Web Vitals: LCP, INP &amp; CLS Thresholds (2026)</title>
+<meta name="description"
+  content="The exact LCP, INP, and CLS pass thresholds Google uses at the p75 field
+  percentile — plus how to check yours in 2 minutes.">
+```
+
+## 2. Schema / JSON-LD — ship the live types, never the dead ones
+
+Pick from the live column. Shipping a deprecated type is **dead markup**: it renders
+nothing and signals carelessness. The single highest-value guard in this skill is *not
+emitting a dead `@type`*.
+
+| Page type | Ship this JSON-LD | Why |
+|---|---|---|
+| Article / blog post | `Article` or `BlogPosting` — `headline`, `author`, `datePublished`, `dateModified`, `image` | Still eligible; `dateModified` is a freshness signal |
+| Product page | `Product` + `Offer` — `price`, `priceCurrency`, `availability`, optional `aggregateRating` | Live rich result |
+| Any page | `BreadcrumbList`, `Organization` / `WebSite` | Entity signals AI engines corroborate against |
+| FAQ section | **NOTHING for rich results** | FAQ rich results restricted to health/gov since late 2023; full deprecation **May 7 2026**, API removed Aug 2026. Keep Q&A as visible prose. |
+| Course / event-announcement / claim-review / salary / vehicle / learning-video | **NOTHING** | Retired **June 2025** — dead markup |
+
+Validate every block with the **Rich Results Test** and the **Schema Markup Validator
+(validator.schema.org)** before shipping. Copy-paste blocks with required-vs-recommended
+props and the full dead-types list are in
+[`references/schema-recipes.md`](references/schema-recipes.md).
+
+## 3. GEO pass — the measured citation levers
+
+The Princeton GEO study quantified what moves source content into AI answers. Apply these
+as concrete edits — they are not vibes, they are deltas measured on GEO-bench:
+
+- **Cited sources inline** (~+30% visibility) — link to authoritative origins for claims.
+- **Statistics** (~+32%) — one verifiable number per ~150–200 words, with its source.
+- **Quotations** (~+41%, the biggest lever) — quote a named expert or primary document.
+- **Fluency** (~+28%) — clean, readable prose. **Keyword stuffing did NOT help** in the
+  study; stuffing is pure cost.
+- **Freshness** — update real content *and* `dateModified`. Recency is a genuine AI-source
+  signal; a stale-looking page is corroborated less.
+- **Entity corroboration** — earn mentions on independent authoritative domains so AI
+  engines can cross-check who you are. This is the off-page half of GEO.
+
+Why it is no longer optional: a citation in an AI Overview can lift CTR 80%+, while AI
+Overviews are *compressing* classic organic CTR (Gartner projected ~25% organic-traffic
+decline by 2026). GEO runs *alongside* on-page SEO, never instead of it.
+
+**Myth callout — `llms.txt`.** Do not sell `llms.txt` as a ranking or AI-visibility
+signal. Google (Illyes, Mueller) does not use it and likened it to the long-ignored
+keywords meta tag; the major AI crawlers overwhelmingly skip `/llms.txt` and crawl HTML
+directly. AI access is governed by **robots.txt + real on-page authority**, not a root
+text file. Spending effort there is the modern keyword-meta-tag mistake.
+
+## 4. Technical audit — measure, flag, hand off
+
+**Core Web Vitals** (thresholds unchanged since INP replaced FID in March 2024):
+
+| Metric | Good (p75) |
+|---|---|
+| LCP | ≤ 2.5s |
+| INP | ≤ 200ms |
+| CLS | ≤ 0.1 |
+
+A URL **passes only if ≥75% of real visits hit "good" at the p75 of the CrUX field
+dataset.** Lab tools (Lighthouse) *estimate*; **CrUX / field decides.** A green Lighthouse
+score with a failing CrUX p75 is still a fail. **Measure and flag here — hand the actual
+render-path fix to `performance`** (and to [`../nextjs/SKILL.md`](../nextjs/SKILL.md) for
+App Router apps). Do not guess at LCP fixes in this skill.
+
+**Crawlability / indexability:** correct `<link rel="canonical">`, a referenced
+`sitemap.xml`, no stray `noindex`/`Disallow` on the page you want ranked, robots.txt
+present and intentional.
+
+**AI-bot access (per purpose):** robots.txt controls AI crawlers separately by purpose,
+and the choice has visibility consequences.
+
+- Blocking the *search* bot (`OAI-SearchBot`, `PerplexityBot`, `Claude-SearchBot`)
+  **removes you from that engine's citations.** Block only the *training* bot
+  (`GPTBot`, `ClaudeBot`, `CCBot`) if you want citations but not training-corpus use.
+- `Google-Extended` opts out of Google AI *training* — but **AI Overviews are served by
+  ordinary Googlebot, so you cannot opt out of AI Overviews without leaving Search.**
+
+The full 2026 user-agent table and robots.txt templates are in
+[`references/ai-crawler-control.md`](references/ai-crawler-control.md).
+
+## 5. The audit checklist (the artifact you emit)
+
+Emit this as a pass/fail list per page. This is what `scripts/verify.sh` lints.
+
+- [ ] `<title>` ≤ 60 chars, query near front; meta description ≤ 160.
+- [ ] Exactly one `<h1>`; H2/H3 hierarchy matches intent.
+- [ ] Primary query answered in the first ~40–200 words.
+- [ ] JSON-LD: valid JSON, has `@context` + `@type`, **no dead type**, validates in Rich
+      Results Test.
+- [ ] ≥2 GEO levers present (statistic / quotation / cited source / fluency).
+- [ ] CWV stated at p75 field thresholds, with pass/fail + handoff note if failing.
+- [ ] canonical + sitemap correct; no accidental `noindex`.
+- [ ] robots.txt does not block a *search* bot while the page claims AI visibility.
+
+## Anti-patterns
+
+| Anti-pattern | Why it fails | Do instead |
+|---|---|---|
+| Shipping FAQPage / Course / Vehicle JSON-LD for rich results | Deprecated (FAQ dead May 2026; seven types retired June 2025) — dead markup | Article/Product/Breadcrumb; keep Q&A as visible prose |
+| Treating GEO as a separate job from Google SEO | AI Overviews use the same Googlebot / ranking / E-E-A-T | One audited page serves both surfaces |
+| Adding `llms.txt` to "rank in AI" | Google ignores it; AI bots skip it | Fix robots.txt + earn real authority/citations |
+| Burying the answer under intro fluff | Snippets & AI extraction need answer-first | Answer the query in the first ~40–200 words |
+| Keyword stuffing for AI citation | Princeton: stuffing didn't help; quotes/stats/cites did | Add statistics, quotations, cited sources |
+| Blocking OAI-SearchBot/PerplexityBot, then expecting AI citations | Search bots feed the citations | Block only training bots if you want citations |
+| Calling CWV "passed" off a green Lighthouse score | CrUX p75 field decides, not the lab | Read CrUX / field at p75; lab only estimates |
+| Debugging an LCP regression inside this skill | Render-path work belongs to performance | Flag the fail, hand to `performance` |
+
+## Handoff
+
+When the audit hits a boundary, name the owner and stop:
+
+- **Keywords, clusters, briefs, "what should we write?"** → `content-engine`.
+- **The actual article/body draft** → [`../article-writing/SKILL.md`](../article-writing/SKILL.md).
+- **Persuasive value-prop / CTA copy** → `landing-copy`.
+- **Fixing a CWV fail (render, hydration, bundle)** → `performance`.
+- **Wiring `generateMetadata` / sitemap route / app-router SEO** → [`../nextjs/SKILL.md`](../nextjs/SKILL.md).
+- **WAI-ARIA / contrast / keyboard a11y** → `accessibility`.
+- **Measuring traffic / GA4 / Search Console dashboards after** → `analytics`.

package/skills/seo-geo/evals/README.md

@@ -0,0 +1,14 @@
+# Evals — seo-geo
+
+`cases.yaml` drives two kinds of check. The `should_trigger` / `should_not_trigger`
+prompts are routing assertions: run them through the repo's eval harness against this
+skill's `description` + triggers and confirm the router selects `seo-geo` for the trigger
+set (including the non-obvious GPTBot/robots and dead-FAQ-schema cases and the
+Spanish/Catalan prompts) and prefers the named sibling for each near-miss
+(`content-engine`, `article-writing`, `performance`, `nextjs`, `accessibility`). The
+`capability` case is graded manually or by an LLM judge against its `must_include` rubric:
+feed the scenario to an agent loaded with this skill and check the produced audit hits
+every rubric line (length-bounded title/meta, answer-first note, valid Article JSON-LD
+with no dead type, ≥2 Princeton GEO levers, CWV stated at p75 with a performance handoff,
+the AI-bot/robots note, and an explicit handoff statement). There is no automated scorer
+in this folder; routing is harness-run, capability is rubric-judged.

package/skills/seo-geo/evals/cases.yaml

@@ -0,0 +1,45 @@
+skill: seo-geo
+
+should_trigger:
+  - prompt: "Audit this blog post for SEO and tell me why it isn't ranking."
+    why: "Core on-page audit of an existing artifact — title/meta/H1/intent/schema/CWV — squarely this skill."
+  - prompt: "We want to get cited in ChatGPT and Perplexity answers. What do we change on the page?"
+    why: "GEO core: the Princeton citation levers + AI-bot access, applied to a concrete page."
+  - prompt: "Write the JSON-LD schema for this product page."
+    why: "Structured data — Product + Offer, with the dead-type guard. The skill emits the block."
+  - prompt: "Should we block GPTBot in our robots.txt, or will that kill our AI visibility?"
+    why: "Non-obvious: looks like a generic devops/robots question, but the training-vs-search-bot trade-off and its citation consequence is exactly seo-geo's AI-crawler-control territory."
+  - prompt: "Audita el SEO de esta página y dime si pasamos Core Web Vitals."
+    why: "Spanish; core technical + on-page audit including the CWV p75 thresholds."
+  - prompt: "Afegeix schema a aquest article del blog perquè surti com a rich result."
+    why: "Catalan; structured-data request for an Article page, with the ship-live-not-dead-types decision."
+  - prompt: "Our page has FAQ schema but we get no rich result anymore — is the markup broken?"
+    why: "Non-obvious: the markup isn't broken, FAQPage rich results are deprecated (health/gov only since 2023, gone May 2026). The skill knows the type is dead and what to do instead."
+
+should_not_trigger:
+  - prompt: "Do keyword research and build a topic cluster plan for our blog."
+    route_to: "content-engine"
+    why: "Upstream strategy — deciding what to write and which terms to target — not optimizing an existing page."
+  - prompt: "Write the actual 1500-word article for this topic."
+    route_to: "article-writing"
+    why: "Long-form body draft; seo-geo optimizes a page, it does not author the prose."
+  - prompt: "Our LCP is 5s — profile the React render and fix the slow hydration."
+    route_to: "performance"
+    why: "Render-path / hydration debugging; seo-geo only measures CWV at p75 and flags, then hands off."
+  - prompt: "Wire up Next.js generateMetadata and the app-router sitemap.ts route."
+    route_to: "nextjs"
+    why: "Framework plumbing of the metadata API; seo-geo decides the meta values, nextjs wires them."
+  - prompt: "Audit our checkout flow for keyboard navigation and color contrast."
+    route_to: "accessibility"
+    why: "WAI-ARIA / contrast / keyboard a11y, a different audit discipline than search visibility."
+
+capability:
+  - scenario: "Given a real article URL (a title, an intro paragraph, a few H2s, no JSON-LD, and unknown Core Web Vitals), produce an SEO+GEO audit with concrete fixes."
+    must_include:
+      - "A rewritten <title> ≤60 chars and meta description ≤160, with the primary query near the front."
+      - "An answer-first rewrite note: the primary query is answered in the first ~40–200 words."
+      - "Valid Article/BlogPosting JSON-LD with required props (headline, author, datePublished, image) and NO dead schema type (no FAQPage/Course/etc.)."
+      - "At least two GEO levers applied (statistic, quotation, cited source, or fluency) tied explicitly to the Princeton findings, with the keyword-stuffing-doesn't-help caveat."
+      - "Core Web Vitals stated as p75 field thresholds (LCP ≤2.5s, INP ≤200ms, CLS ≤0.1) with a measure-then-hand-to-performance note and the lab-vs-CrUX distinction."
+      - "A robots.txt note on AI-bot access: don't block OAI-SearchBot/PerplexityBot if citations are wanted; llms.txt is not a visibility signal."
+      - "An explicit handoff statement naming the owners (content-engine / article-writing / performance / nextjs) for out-of-scope work."

package/skills/seo-geo/references/ai-crawler-control.md

@@ -0,0 +1,104 @@
+# AI-crawler control (robots.txt, 2026)
+
+AI access to your site is governed by **robots.txt + real on-page authority** — not by
+`llms.txt` (Google ignores it; AI bots skip it). The lever that matters is *which
+user-agent you allow*, because vendors split their crawlers **by purpose**, and blocking
+the wrong one silently removes you from an engine's citations.
+
+## The purpose split
+
+| User-agent | Vendor | Purpose | Block effect |
+|---|---|---|---|
+| `GPTBot` | OpenAI | Training corpus | Out of training; **still citable** |
+| `OAI-SearchBot` | OpenAI | ChatGPT Search index | **Out of ChatGPT Search citations** |
+| `ChatGPT-User` | OpenAI | Live user-triggered fetch | No live fetch on user action |
+| `ClaudeBot` | Anthropic | Training corpus | Out of training; still citable |
+| `Claude-SearchBot` | Anthropic | Search index | Out of Claude search citations |
+| `Claude-User` | Anthropic | Live user-triggered fetch | No live fetch |
+| `PerplexityBot` | Perplexity | Search index | **Out of Perplexity citations** |
+| `Google-Extended` | Google | Gemini/Vertex training opt-out | Out of Google AI training |
+| `Googlebot` | Google | Search crawl + **AI Overviews** | Out of Search entirely (see caveat) |
+| `Applebot-Extended` | Apple | Apple AI training opt-out | Out of Apple AI training |
+| `CCBot` | Common Crawl | Open dataset (feeds many models) | Out of Common Crawl |
+| `meta-externalagent` | Meta | Meta AI crawl | Out of Meta AI |
+
+## Caveats that bite
+
+- **Google AI Overviews ride ordinary Googlebot.** `Google-Extended` opts you out of
+  Gemini *training* only. There is **no way to opt out of AI Overviews without blocking
+  Googlebot — i.e. leaving Google Search.** Do not promise a client "AI Overviews opt-out."
+- **Search bot ≠ training bot.** "We blocked GPTBot" does **not** remove you from ChatGPT
+  Search citations — that is `OAI-SearchBot`. People routinely block the wrong one.
+- **Perplexity circumvention.** Cloudflare reported (Aug 2025) that Perplexity fetched
+  pages via undeclared agents after being disallowed. robots.txt is an honor-system
+  directive; enforce hard blocks with a WAF/firewall, not just robots.txt, if you truly
+  must keep a bot out.
+
+## Template — "cite me, but don't train on me"
+
+Keep search/citation bots in; keep training bots out.
+
+```text
+# robots.txt — citation-eligible, training opted out
+
+User-agent: GPTBot
+Disallow: /
+
+User-agent: ClaudeBot
+Disallow: /
+
+User-agent: CCBot
+Disallow: /
+
+User-agent: Google-Extended
+Disallow: /
+
+# Search/citation bots: ALLOWED (no Disallow = allowed)
+User-agent: OAI-SearchBot
+Allow: /
+
+User-agent: PerplexityBot
+Allow: /
+
+User-agent: Claude-SearchBot
+Allow: /
+
+User-agent: Googlebot
+Allow: /
+
+Sitemap: https://example.com/sitemap.xml
+```
+
+## Template — "block all AI"
+
+Accepts loss of AI citations. Note this does NOT block Google AI Overviews (that needs
+blocking Googlebot, which leaves Search).
+
+```text
+# robots.txt — block AI training AND AI search
+
+User-agent: GPTBot
+User-agent: OAI-SearchBot
+User-agent: ChatGPT-User
+User-agent: ClaudeBot
+User-agent: Claude-SearchBot
+User-agent: Claude-User
+User-agent: PerplexityBot
+User-agent: CCBot
+User-agent: Google-Extended
+User-agent: Applebot-Extended
+User-agent: meta-externalagent
+Disallow: /
+
+# Googlebot left ALLOWED — blocking it would remove you from Search.
+User-agent: Googlebot
+Allow: /
+
+Sitemap: https://example.com/sitemap.xml
+```
+
+## Audit check
+
+If the brief says "we want AI citations," confirm robots.txt does **not** `Disallow: /`
+under `OAI-SearchBot`, `PerplexityBot`, `Claude-SearchBot`, or `Googlebot`. A self-blocking
+robots.txt is the most common reason a "GEO-optimized" page earns zero citations.

package/skills/seo-geo/references/schema-recipes.md

@@ -0,0 +1,130 @@
+# Schema / JSON-LD recipes
+
+Copy-paste blocks for the structured-data types that **still earn rich results in 2026**,
+plus the explicit list of types Google has **killed** (do not ship these). Validate every
+block with the **Rich Results Test** and the **Schema Markup Validator
+(validator.schema.org)** before it goes live.
+
+All blocks go in a `<script type="application/ld+json">` in `<head>` or at the end of
+`<body>`. JSON-LD must be valid JSON (double-quoted keys, no trailing commas) and carry
+`@context` + `@type`.
+
+## Article / BlogPosting
+
+Use for any editorial page. `dateModified` doubles as a freshness signal for AI sources.
+
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "BlogPosting",
+  "headline": "Core Web Vitals: LCP, INP & CLS Thresholds (2026)",
+  "image": ["https://example.com/img/cwv-cover.png"],
+  "author": {
+    "@type": "Person",
+    "name": "Jane Doe",
+    "url": "https://example.com/authors/jane-doe"
+  },
+  "publisher": {
+    "@type": "Organization",
+    "name": "Example",
+    "logo": { "@type": "ImageObject", "url": "https://example.com/logo.png" }
+  },
+  "datePublished": "2026-01-15T08:00:00+01:00",
+  "dateModified": "2026-06-02T09:30:00+02:00",
+  "mainEntityOfPage": "https://example.com/blog/core-web-vitals"
+}
+```
+
+- **Required for the rich result:** `headline`, `image`, `datePublished`, `author`.
+- **Strongly recommended:** `dateModified`, `publisher`, `mainEntityOfPage`.
+
+## Product + Offer
+
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "Product",
+  "name": "Acme Wireless Headphones",
+  "image": ["https://example.com/img/headphones.png"],
+  "description": "Over-ear ANC headphones, 40h battery.",
+  "sku": "ACME-WH-200",
+  "brand": { "@type": "Brand", "name": "Acme" },
+  "offers": {
+    "@type": "Offer",
+    "url": "https://example.com/p/acme-wh-200",
+    "priceCurrency": "EUR",
+    "price": "199.00",
+    "availability": "https://schema.org/InStock"
+  },
+  "aggregateRating": {
+    "@type": "AggregateRating",
+    "ratingValue": "4.6",
+    "reviewCount": "318"
+  }
+}
+```
+
+- **Required for the offer result:** `name`, `image`, `offers` with `price` +
+  `priceCurrency` + `availability`.
+- **Recommended:** `aggregateRating`/`review` (only with genuine on-page reviews), `brand`, `sku`.
+
+## BreadcrumbList
+
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "BreadcrumbList",
+  "itemListElement": [
+    { "@type": "ListItem", "position": 1, "name": "Blog", "item": "https://example.com/blog" },
+    { "@type": "ListItem", "position": 2, "name": "Core Web Vitals", "item": "https://example.com/blog/core-web-vitals" }
+  ]
+}
+```
+
+## Organization / WebSite (entity signals)
+
+These corroborate *who you are* for both Google's Knowledge Graph and AI engines.
+
+```json
+{
+  "@context": "https://schema.org",
+  "@type": "Organization",
+  "name": "Example",
+  "url": "https://example.com",
+  "logo": "https://example.com/logo.png",
+  "sameAs": [
+    "https://www.linkedin.com/company/example",
+    "https://github.com/example"
+  ]
+}
+```
+
+## DEAD types — never ship these for rich results
+
+Shipping any of these renders nothing and is dead markup. Verify.sh fails the build if it
+finds one.
+
+| Dead `@type` | Status |
+|---|---|
+| `FAQPage` | Rich results restricted to health/gov since late 2023; **full deprecation May 7 2026**, API removed Aug 2026 |
+| `Course` (Course Info rich result) | Retired **June 2025** |
+| `ClaimReview` | Retired June 2025 |
+| `SpecialAnnouncement` | Retired June 2025 |
+| `VehicleListing` | Retired June 2025 |
+| `EstimatedSalary` | Retired June 2025 |
+| `LearningVideo` | Retired June 2025 |
+| Book Actions | Retired June 2025 |
+
+**Do instead:** for FAQ content, keep the questions and answers as **visible prose** on
+the page (answer-first, the GEO levers still apply) — you simply do not wrap them in
+FAQPage markup. The content stays; only the dead schema goes.
+
+`HowTo` is also no longer eligible for the rich result on most surfaces — treat it as
+display-only, not a visibility play.
+
+## Validation step (always)
+
+1. Paste the rendered page into the **Rich Results Test** (search.google.com/test/rich-results)
+   — confirms eligibility for a specific rich result.
+2. Paste into **validator.schema.org** — confirms the markup is well-formed schema.org.
+3. Reject the block if either reports a dead/unsupported type or a missing required prop.