SEO + GEO + AEO Playbook

You are implementing three surfaces at once. They share one content system. They do not share one metric.

Surface	What wins	Primary index
SEO	Rank + CTR in classic SERPs	Google
GEO	Get quoted in ChatGPT, Perplexity, Claude, Gemini	Bing (ChatGPT) + Google (Perplexity/Gemini)
AEO	Be the cited source in AI Overviews / AI Mode / snippets	Google

Non-negotiable thesis: GEO is additive to SEO, not a replacement. ~80% of GEO is excellent fundamental SEO. Ranking top-10 no longer guarantees an AI citation — you also need extractability. Organic still sends far more raw traffic than all AI engines combined; AI-referred sessions convert much harder. Build both.

Never fabricate stats, quotes, experts, or studies. Engines treat that as adversarial. Find real sources or cut the claim. See content-system.md.

Step 0 — Bind the project

Do not start writing pages until these are named. Infer from the repo; ask only for gaps.

Canonical host (https, one host; apex 301s to it)
Primary conversion URL (the tool/signup/demo the content exists to feed)
Entity facts (legal name, founding, HQ, category, 1-sentence what-it-does) — identical everywhere
Locked terminology (words that must never appear; canonical names for the product)
Money query (the one head term) + cluster queries (long-tail that feed it)
Author (real Person with credentials + sameAs URLs)

Write them into a short search-visibility.md at the project root (or update it). Every later page inherits this.

Step 1 — Pick the job

Infer from the user. If ambiguous, ask once.

Signal	Job	Read
"audit the site / why aren't we ranking / GSC"	Audit	technical-stack.md then entity-and-measurement.md
"write a post / landing / cluster"	Content	content-system.md
"schema, robots, sitemap, IndexNow, metadata"	Technical	technical-stack.md
"Wikidata, Reddit, YouTube, citations, GA4 AI channel"	Entity + measure	entity-and-measurement.md
"full program / 90-day plan / playbook"	Program	all three, then the 90-day sequence below

Step 2 — The content law (every public URL)

Every indexable page — blog, tool, landing, comparison — follows this skeleton. Skip a layer and citation rate collapses.

1. ANSWER FIRST          Direct answer in the first screenful (≤ first 30% of the page).
                         Bold the verdict. 2–4 sentences. 44% of LLM citations come from here.
2. EXTRACTION TABLE      Exactly one self-contained comparison/data table near the top.
                         An LLM must be able to quote the table with no surrounding prose.
3. PRIMARY CTA           Early, specific, to the conversion URL. Repeat at the close.
4. QUESTION H2s          Each H2 is a real query. First sentence under it is a complete answer.
                         Then elaborate. One idea per section. Sections must stand alone.
5. EVIDENCE DENSITY      ≥1 verifiable number per ~100 words. Inline primary citations.
                         ≥1 attributable expert/source quote on guides.
6. DIFFERENTIATION       Short section: why THIS product, in locked terminology, with operator-locked proof only.
7. FAQ                   3–6 questions people actually type or ask a chatbot. Answers 2–4 sentences, quotable verbatim.
8. SOURCES               Bulleted list of every cited URL with a one-line parenthetical.

Title tag: primary query front-loaded, ≤60 characters, hook (number, audience, or concrete outcome) — not the bare keyword. Meta description: 150–160 characters, query + benefit + CTA. H1 matches title intent. Canonical is self. OG/Twitter complete (1200×630).

Full rules, anti-patterns, and QA gate: content-system.md.

Step 3 — Technical foundation (ship once, keep green)

Do these in order. Details and copy-paste artifacts: technical-stack.md.

One canonical host + HTTPS + trailing-slash policy + rel=canonical on every public page
robots.txt — allow Googlebot, Bingbot, and AI retrieval bots (GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, Claude-SearchBot, ClaudeBot, Google-Extended, Applebot-Extended). Disallow app/auth/admin. Block abusive scrapers (Bytespider, CCBot). Sitemap URL declared.
XML sitemap of public URLs only; submit to Google Search Console and Bing Webmaster (ChatGPT reads Bing)
IndexNow on publish/update — Bing+Yandex instant ping. Google ignores it; do it anyway for GEO
JSON-LD @graph with ≥3 types per URL. Schema must match visible content. Keep FAQPage even after Google killed FAQ rich results — it is now an AI-extraction signal
Person author on articles/guides (jobTitle, url, sameAs). Organization sameAs = Wikidata + LinkedIn + Crunchbase + socials
SSR or prerender the answer, table, and FAQ. JS-only main content is invisible to many AI crawlers
llms.txt (and optional llms.txt) at site root — cheap content map. Do not invest maintenance beyond keeping URLs accurate; no major engine uses it for retrieval. robots.txt is the file that matters
Core Web Vitals green on mobile (LCP < 2.5s, INP < 200ms, CLS < 0.1). Title/H1 in first paint
hreflang only if you truly have translated equivalents

Step 4 — Information architecture

Pillar + cluster. One pillar per money query. Cluster posts answer fan-out questions and link up with descriptive anchors (never "click here").
Tool pages are content pages. ChatGPT over-cites product/tool URLs vs Perplexity. Put the answer, method, table, FAQ, and citations on the tool URL — not only a widget.
Predict-X-from-Y / comparison / how-to pages own the long tail. Each must have non-trivial synthesis (unique table, unique sources, unique advice). Thin programmatic templates get a weakest-link domain penalty.
No cannibalization. One URL owns one primary query. Others 301, canonicalize, or differentiate.
Internal links: every new page gets 2–3 inbound links from related live pages in the same cluster.

Step 5 — Entity layer (the GEO moat)

On-page extractability is necessary and not sufficient. Engines cite entities they recognize.

Priority order (full steps in entity-and-measurement.md):

Identical NAP + one-sentence description on site JSON-LD, Wikidata, LinkedIn, Crunchbase, Product Hunt, G2/Capterra as relevant
Wikidata item with official website + sameAs mirrored in Organization schema
Wikipedia only after independent press — do not create a promo stub
Genuine Reddit/community answers (no astroturf). Perplexity and AI Overviews overweight Reddit
2–3 YouTube explainers/comparisons (Wikipedia + YouTube + Reddit ≈ a large share of AI-cited media)
Digital PR for unlinked brand mentions in trade press

Step 6 — Measure or you are guessing

Only a small minority of teams track AI search. Install this in week 1.

GA4 — native "AI Assistants" channel, or a custom channel above Referral matching chatgpt.com|chat.openai.com|perplexity.ai|claude.ai|gemini.google.com|copilot.microsoft.com. Custom groups do not backfill — create immediately. AI Overviews have no separate referrer; use GSC.
GSC — domain property; AI Mode / AI Overview filter when available; question-query regex
Bing Webmaster — sitemap indexed (ChatGPT path)
Share of voice — monthly: 20 money prompts across ChatGPT, Perplexity, Gemini, AI Overviews. Record cited / not cited / competitor cited
KPIs: non-branded clicks + impressions, AI-referral sessions + conversion, citation SoV on the 20 prompts, Bing coverage

90-day sequence (when building a program)

Days	Work
1–30	Canonical + robots AI allowlist + dual sitemaps + IndexNow + GA4 AI channel + GSC/Bing verify. Content-engineer the money URL and top 5 pages (answer-first, table, citations, FAQ, Person author, 3+ schema types).
31–60	Pillar + first cluster (8–12 URLs). Wikidata + listings `sameAs`. Author bylines. Refresh any page older than 90 days that still ranks.
61–90	Reddit/YouTube presence. Digital PR. Comparison/alternative pages. Citation SoV baseline → second measurement. Kill or 301 thin URLs.

Quality gate (nothing ships without this)

First 150 words answer the primary query without throat-clearing
One extraction table; every cell understandable alone
Every H2 is a question; first sentence under it is a complete answer
Every load-bearing number has a fetched, clickable primary source
Zero fabricated quotes/stats; zero keyword stuffing (target term < 1% of words)
FAQ answers are verbatim-quotable and match visible on-page FAQ
JSON-LD @graph has ≥3 relevant types; URLs absolute; dates ISO-8601; author is Person
Canonical, OG, Twitter, sitemap entry, IndexNow ping
AI bots allowed; private routes disallowed
2–3 internal links in; 2–3 related links out
Product claims use only operator-locked numbers
Mobile CWV not regressing

Anti-patterns (hard fails)

Keyword stuffing, filler intros ("In today's landscape…"), unsourced superlatives
FAQ schema for questions not visible on the page
Blocking GPTBot / PerplexityBot / ClaudeBot / Google-Extended
Thin city/template pages at scale
Investing weeks in llms.txt instead of citability
Treating AI Overviews as a GA4 "AI Search" channel (they are organic)
Naming competitor formulas/trademarks the product forbids
Publishing unverified product stats

Coordination

Sitewide GSC crawl + striking-distance list: use the project's seo-analysis skill if present, then apply this playbook to the fixes
Single URL rewrite: apply content-system.md + schema from technical-stack.md
Meta-only CTR work: title/description formulas in content-system.md
Do not spawn a separate "GEO voice" that contradicts the product terminology skill

seo-geo-aeo

Files4

SKILL.md.cursor/skills/seo-geo-aeo/SKILL.md