GEO and AEO: how to get cited in AI Overviews and AI Mode

Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO) describe getting cited inside AI answers like Google's AI Overviews and AI Mode. There is no separate GEO ranking system: to be eligible, a page needs only to be indexed and able to show in Search with a snippet. Everything past that is ordinary SEO done well.

Eligibility: nothing special to do

Google is direct about this. In its own words, "optimizing for generative AI search is optimizing for the search experience, and thus still SEO," and "there are no additional requirements to appear in AI Overviews or AI Mode." The AI features run on top of core Search ranking and quality systems, and your E-E-A-T signals are processed by the same model.

Google has explicitly debunked several things sold as GEO requirements. You do not need any of these:

• Machine-readable, AI-only, or Markdown files — "Google Search ignores them."
• Special schema.org markup "for AI."
• Content chunking.
• An AI-only writing style.
• Inauthentic "mentions."

And a warning: producing separate content for every possible query variation risks tripping Google's scaled content abuse policy. The impulse to mass-produce AI-targeted pages is the fastest way to get demoted.

The llms.txt reality check

llms.txt is heavily promoted as an AI-SEO must-have. The evidence says otherwise. Roughly 97% of llms.txt files received zero requests (Ahrefs, May 2026), and of the requests that did land, AI retrieval bots were about 1.1% — the top requesters were SEO audit tools, not AI answer engines. Google does not use it. John Mueller compared it (June 2026) to the long-dead keywords meta tag, noting "none of the AI systems use it" and that it "lacks a discovery mechanism."

The one legitimate use is developer-product documentation consumed by coding assistants such as Cursor and Continue. Do not build llms.txt expecting citations or traffic. If you want a single actionable rule for agentic optimization, Google's is simpler: do not block agents.

How links actually surface in AI answers

Two mechanisms decide what an AI answer cites, and both sit on top of traditional Search.

Retrieval-augmented generation (RAG): core Search ranking retrieves relevant pages first, then the AI presents prominent clickable links from that set. If you do not rank, you are not in the candidate pool.

Query fan-out: the system issues several related queries at once and pulls a wider set of links than a single SERP would. AI Overviews and AI Mode still fan out to traditional search underneath, so organic ranking remains the substrate. This is why "you must already rank" holds — but ranking alone is no longer sufficient. As of mid-2026, only 38% of AI Overview citations come from the top-10 organic results (down from 76% in July 2025); 62% come from outside the top-10 (Ahrefs, 2026). Query fan-out is the driver: the model issues multiple related queries and draws from a far wider candidate pool than any single SERP.

What measurably earns citations

You cannot game the model, but research has begun to isolate what correlates with citation. The most rigorous published analysis (Zyppy/Shepard meta-analysis of 54 experiments, May 2026) found the strongest measurable correlate of AI citation is off-site brand mentions (r=0.664) — roughly 3x stronger than backlinks alone (r=0.218). Branded anchor text (r=0.527), brand search volume (r=0.392), and domain rating (r=0.326) follow. Building a recognizable brand that earns editorial coverage and unlinked mentions is the highest-leverage lever the data supports.

The following on-page patterns also correlate with being cited, per the Princeton GEO study and community measurement.

• Answer-first capsule. Open every page, and every major H2, with a 40–60 word paragraph that directly answers the query. AI extracts these near-verbatim, and around 44–55% of AI citations come from the first 30% of a page.
• Credibility elements. Adding quotations lifts citation likelihood ~41%, statistics ~32%, and inline citations ~30%.
• Write definitively. Cited text is about twice as likely to use direct language. Cut hedging where the facts are settled.
• Format to intent. Listicles win commercial and comparison queries (a large share of LLM citations are numbered "Top-N" lists); long-form wins informational queries; clean product pages win transactional ones.
• Word count is not a factor. The correlation is near zero (~0.04). Around 1,000–1,500 words is fine — do not pad.
• Earn media. Journalism and PR are disproportionately strong citation sources (roughly a quarter of cited sources), and freshness matters.

Being cited is increasingly "the new position one": brands cited inside an AI Overview earn about 35% more organic clicks, a rare bright spot in an era where roughly 68% of US Google searches now end without a click.

Let the right AI crawlers in

Robots policy for AI splits into three functions, and mixing them up is the most common mistake.

• Search/retrieval bots — OAI-SearchBot, Claude-SearchBot, PerplexityBot. Allow these if you want citations.
• Training bots — GPTBot, ClaudeBot, CCBot, Bytespider, and the Google-Extended / Applebot-Extended opt-out tokens. Blocking these keeps you out of training data only. It does not remove you from live AI answers.
• User-triggered fetchers — ChatGPT-User, Claude-User, Perplexity-User. Vendor policies are inconsistent.

The critical caveat: blocking training bots does not remove you from live AI answers, because those come from the search bots. And Google-Extended is an opt-out token, not a crawler — disallowing it opts out of Gemini training, not Search ranking or AI Overview eligibility. To opt out of AI Overviews without losing normal Search, use the Search Console toggle instead.

How Crawlinx helps

Crawlinx's Agent-readiness checks target the one honest lever Google actually endorses — do not block agents, and keep pages indexable and well-structured — rather than the debunked tactics.

• robots.blocked — catches pages blocked from crawlers, including the AI retrieval bots that drive citations.
• index.noindex — flags pages that cannot be indexed, which makes them ineligible for AI answers by definition.
• content.thin — surfaces pages too thin to be retrieved or cited.
• schema.none — notes missing structured data on pages where valid markup supports entity understanding (see the structured data guide).
• agent.no_main_landmark — checks for a <main> landmark, which helps agents and assistive tech locate primary content.
• agent.llms_txt_missing — reports llms.txt presence as an informational signal for developer-doc use cases, not as a ranking or citation requirement.

Takeaway

Treat GEO and AEO as SEO with an answer-first structure: rank in organic Search, open pages and sections with a direct 40–60 word answer, add quotations and statistics, and let the AI retrieval bots crawl you. Skip llms.txt and AI-specific schema — Google does not use them, and mass-producing AI-targeted pages risks a scaled-content penalty.

Related reading: Core Web Vitals and structured data.