awesome-geo

English | 한국어 | 中文 | 日本語 | Español | Français

A curated, source-first map of Generative Engine Optimization (GEO), Answer Engine Optimization (AEO), and AI search — built on official engine documentation, not vendor hype.

The uncomfortable thesis of this list: by the search engines' own official docs, GEO/AEO is not a secret new optimization discipline. For Google Search it is foundational SEO applied to new answer surfaces (AI Overviews, AI Mode). Most other AI search engines (ChatGPT, Perplexity, Claude) officially share the same base — crawlable, indexable, trustworthy, fresh — and publish no separate "rewrite your content for AI" guidance. The one engine that does take a visibly different official stance is Microsoft Bing/Copilot. The real difference between engines is not how you write but the retrieval backend each one reads from.

Every claim here links to a primary source. Marketing blog posts are kept in a clearly-labeled secondary section and are never promoted to fact.

• The Core Argument
• Terminology Map
• Google — Official Docs
• Measuring AI Search Performance
• Per-Engine Official Stances
• llms.txt — What It Actually Is
• Social / SNS Surfaces
• Industry & Academic References
• What to Ignore
• Contributing

AEO (Answer Engine Optimization) and GEO (Generative Engine Optimization) are real, widely-used industry terms for getting a brand or content cited in AI answers — AI Overviews, AI Mode, ChatGPT, Perplexity, voice search. GEO was formalized academically by the 2023 paper GEO: Generative Engine Optimization (KDD 2024).

The disputed part is not whether the terms exist — it is whether they describe a separate optimization system. Google's own generative-AI optimization guide acknowledges these terms are "common online," then states plainly: for Google Search, you keep applying foundational SEO. No special schema, no llms.txt, no AI-only markdown, no content chunking, no AI-only rewrites are required for AI Overviews / AI Mode.

So this list treats GEO as SEO basics applied to new answer surfaces — with engine-specific caveats documented from primary sources, not invented.

The foundation. If you read nothing else, read the first row.

Core question: Google says GEO/AEO = SEO — do other major AI engines officially claim their engine is different?

Short answer: most share Google's base (crawlable · indexable · trustworthy · fresh) and require no official "secret optimization." The one clearly different official stance is Microsoft Bing/Copilot — Bing states that schema markup helps its LLM understand content, and recommends IndexNow for freshness. And the real cross-engine difference is the retrieval backend (own index vs. third-party search provider), not the writing.

Per-platform primary sources:

• OpenAI / ChatGPT — Publishers & developers FAQ, ChatGPT search, bots / crawlers. OAI-SearchBot (search/product) is separate from GPTBot (training). The official guidance is "don't block it, or use noindex to opt out" — not "rewrite your content for AI."
• Perplexity — crawlers, robots.txt policy. PerplexityBot = search visibility (not training); Perplexity-User = user-triggered fetch. Docs cover technical access only. (Entity-clarity / front-loading / Reddit-citation tips are third-party reverse-engineering, not official — not promoted here.)
• Anthropic / Claude — web search tool, enabling web search. Underlying search relies on a third-party provider (reported: Brave Search); responses carry inline citations and conservatively prefer verifiable primary sources. "Getting picked up by Claude" ≈ "being in the search provider's index as a trustworthy primary source."
• Microsoft Bing / Copilot — AI Performance in Bing Webmaster Tools, Bing/Copilot uses schema for its LLMs. The most opinionated of the set — but Microsoft's Fabrice Canel still stresses "less is more: precision, not volume."

Origin: proposed 2024年09月03日 by Jeremy Howard (fast.ai / Answer.AI co-founder) via an answer.ai post and the llmstxt.org spec. A curated markdown index at site root /llms.txt meant to help an LLM use the site at inference time. It is a one-person community proposal, not a search-consortium standard like schema.org — which is the root cause of the support gap below.

Support status (as of 2026 Q1):

• Google — officially unsupported. John Mueller (2025) and Gary Illyes (Search Central Live 2025-07) stated no AI system currently uses llms.txt and Google has no plans to.
• OpenAI / others — no official announcement that GPTBot/ChatGPT parses llms.txt.
• "Anthropic/Perplexity confirmed support" — some SEO blogs claim this; other reporting says major platforms ignore it and adoption has stalled. No primary confirmation → not promoted to fact here. (Anthropic publishing an llms.txt on its own docs is publisher behavior — it does not mean Claude consumes yours.)

llms.txt remains a community convention, not an official standard.

SNS is less a separate "AEO/GEO play" and more a public web surface indexed by Google Search + brand/profile connection signals.

These corroborate that the terms exist; they are not promoted to Google policy or ranking fact.

Per Google's official position, treat these as red flags:

• "You need an llms.txt for Google Search"
• "AI search needs a special/secret schema"
• "You must chop content into tiny chunks for AI"
• Awkward rewrites written only for AI systems
• Artificial mention farming
• "Guaranteed #1 AI Overview ranking" sales pitches

Safe common denominator across ChatGPT / Perplexity / Claude / others:

• Publicly accessible source text
• Clear author / org / date / evidence
• Canonical, non-duplicate URLs
• Trustworthy external citations
• Up-to-date product / price / policy / FAQ info
• Consistent naming of your brand and key entities

Contributions welcome — but this list has one hard rule: primary source or it doesn't go in.

• Official engine/standards docs → main sections.
• Vendor blogs / analyses → the Industry & Academic References section, clearly labeled, never stated as fact.
• Reverse-engineered "ranking factor" claims without official confirmation are rejected.

See contributing.md. Open a PR or an issue.

CC0

To the extent possible under law, the authors have waived all copyright and related rights to this work.