Generative engine optimization (GEO) is the practice of making your pages easy for AI search engines such as Google AI Overviews, ChatGPT search, Gemini and Perplexity to find, trust and cite in their answers. In practice, GEO is mostly SEO aimed at more queries: get retrieved for the question and its sub-questions, then be the clearest source.
Key takeaways
- AI engines cite pages they first retrieve, so ranking for the query and its sub-questions matters more than any rewriting trick.
- The famous "up to 40%" GEO gains came from a fixed test setup; a peer-reviewed re-test found only 3 of 54 cases significant.
- Google says llms.txt, special schema and "chunking" are not needed for AI Overviews or AI Mode.
- Non-commodity content (your own data, examples, experience) is what Google asks for and what competitors rarely have.
- Measure GEO per keyword: your Google rank plus whether ChatGPT and Gemini cite your domain.
What is generative engine optimization (GEO)?
Generative engine optimization is the work of earning visibility inside AI-generated answers instead of only in a list of blue links. The term comes from a 2023 paper co-authored by researchers at IIT Delhi and Princeton, GEO: Generative Engine Optimization (published at KDD 2024), which defined "generative engines" as systems that retrieve sources and write an answer from them.
You will also see "GEO SEO", "AI search optimization", "answer engine optimization (AEO)" and "LLM SEO". In common usage they describe the same goal with different emphasis, and we break the differences down in GEO vs SEO vs AEO. Google's position is blunt: its guide to optimizing for generative AI features says that optimizing for generative AI search "is optimizing for the search experience, and thus still SEO."
How do AI search engines choose which pages to cite?
AI search engines cite in two steps: they retrieve candidate pages from a search index, then the language model writes an answer from those pages and links some of them. A page that is not retrieved cannot be cited, however well it is written.
Google describes its AI features as grounded in pages from the Search index and built with query fan-out: the model issues several related searches across subtopics before answering. Google's own example is a question about a weedy lawn that fans out into searches about herbicides, chemical-free removal and prevention. For ChatGPT search, OpenAI's crawler is the gate: OpenAI's bot documentation states that sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers."
Citation data fits this picture. In Ahrefs' March 2026 study of 863,000 keyword SERPs and 4 million AI Overview URLs, only 37.9% of cited pages ranked in the top 10 for the exact query; 31.2% ranked 11–100 and 31.0% beyond 100. Ahrefs suggests fan-out may explain much of the drop from about 76% in its July 2025 study, but it also improved its parsing in between, so the comparison is not like-for-like. If that reading is right, a site that ranks for the sub-questions can be cited without ranking for the head term.
What the research says works, and what is myth
The strongest evidence says retrieval and relevance decide citations, while most text "tricks" do little. The studies and official guidance we checked, as of October 2026:
Tactic | What the evidence says | Verdict |
|---|---|---|
Ranking for the query and its sub-questions | In C-SEO Bench (NeurIPS 2025), moving a page to first place in the AI's context gave far larger citation gains than any rewriting method | Highest priority |
Adding statistics, quotes or citations to existing text | Up to 40% gains in the original GEO paper's fixed setup; in C-SEO Bench, statistics lowered rank in 19 of 24 settings | Do it for readers, not as a citation lever |
A short summary at the top | One of only two C-SEO Bench methods with any significant gain, and only for product queries (retail, video games) | Cheap; do it |
Keyword stuffing | Scored below the unoptimized baseline on the GEO paper's main visibility metric; also against Google's spam policies | Harmful |
llms.txt | Google Search ignores it; Ahrefs found 97% of llms.txt files got zero requests in May 2026 | Myth |
Schema markup for AI citations | No special schema needed (Google); Ahrefs saw no major citation uplift after 1,885 pages added JSON-LD | Hygiene, not a lever |
Chunking content into tiny pieces | Google: "There's no requirement to break your content into tiny pieces" | Myth |
Off-site brand mentions | Correlate with AI Overview visibility (0.664) about twice as strongly as Domain Rating (0.326) in Ahrefs' 75,000-brand study; correlation only | Earn real ones |
Plain, predictable wording | AI Overviews cited lower-perplexity text more often in a 2025 preprint; organic rankings showed little effect | Weak signal, free to do |
Why the famous "40% boost" doesn't carry over
The original GEO paper tested rewrites on pages already among the five sources fed to gpt-3.5-turbo. That shows which of five fixed pages gets quoted more; it never measures whether a page gets retrieved. Even there, several tactics reduced visibility for sources already ranked first. The C-SEO Bench re-test across GPT and Claude models found significant gains in only 3 of 54 cases and called most methods "largely ineffective." A 2026 preprint, SAGEO Arena, reports that existing GEO approaches can even hurt retrieval and reranking in a full search pipeline.
Is SEO dead now with AI?
No. Google says its generative AI features "are rooted in our core Search ranking and quality systems," and each AI search engine above retrieves from a search index before it answers. What changes is the unit of work: you compete for the cluster of sub-questions an AI engine fans out to, not one head keyword.
How to optimize for AI search: a 7-step GEO plan
Optimizing for AI search means covering a topic so completely that your pages get retrieved for its sub-questions, then answering each one clearly.
- Map the sub-questions. List the People Also Ask questions, related searches and the headings of the top-ranking pages for your topic. Each real sub-question gets its own clearly headed section or, if the intent differs, its own article in a pillar-and-cluster structure.
- Answer first. Open each section with a one- or two-sentence direct answer, then the detail. Put a short summary at the top of the page. Write each section so it makes sense if quoted alone.
- Add non-commodity content. Google's guide contrasts a generic "7 Tips for First-Time Homebuyers" with a first-hand account of waiving an inspection. Add your own data, worked examples, screenshots and decisions others can't copy.
- Cover the terms that match the query. Use the vocabulary the ranking pages use under their headings, naturally, so retrieval systems see you as on topic. Don't rewrite those terms away for "style."
- Let the right crawlers in. Allow the search crawlers below and put all article text in the server-rendered HTML.
- Earn genuine mentions. Appear in places AI engines read: industry roundups, partner pages, communities, YouTube. Google warns that inauthentic mentions are less helpful than they seem.
- Measure and adjust. Track your Google rank and AI citations per keyword, then improve the pages that rank but aren't cited.
How to do steps 1 and 4 in RankTera: in KW Research, enter your keyword and country. RankTera reads the top-ranking Google pages, extracts the phrases they use under their H2–H6 headings with each phrase's average count (AVG COUNT), and shows the pages' average word count and heading structure, so you can see which sub-topics and terms your draft is missing.
Which AI crawlers should you allow?
Allow every search and user-fetch crawler; training crawlers are a separate business choice. Most AI fetchers don't run JavaScript (Vercel, 2024: none of the major AI crawlers rendered it, with Gemini and Applebot the exceptions), so text that loads client-side may never be seen.
Crawler | Used for | Allow for AI visibility? |
|---|---|---|
Googlebot | Google Search, including AI Overviews and AI Mode | Yes, never block |
Google-Extended | Gemini training and grounding outside Search | Your choice; no effect on Search |
OAI-SearchBot / ChatGPT-User | ChatGPT search and user-requested fetches | Yes |
GPTBot | OpenAI model training | Your choice |
Claude-SearchBot / Claude-User | Claude search indexing and user fetches (Anthropic) | Yes |
ClaudeBot | Anthropic model training | Your choice |
PerplexityBot | Perplexity search results (Perplexity) | Yes |
How do you measure GEO results?
GEO is measured per keyword and per engine: are you cited, and if not, who is? For Google, AI Overview and AI Mode traffic is counted in Search Console's Performance report under the Web search type, and Google's guide points to a Generative AI performance report. For ChatGPT and Gemini, run your keywords with web search on and record whether your domain appears among the cited sources, ideally weekly. Our guide to AI search visibility tracking covers metrics and tools.
One caveat: Google notes that "No third-party tool has access to our internal ranking or AI systems." Every tracker, ours included, observes answers from outside.
Worked example: auditing a GEO cluster
This is an illustrative example, not client data. A company selling invoicing software tracks five keywords around late payments, with their Google rank and whether ChatGPT and Gemini cite its domain.
Keyword | Google rank | ChatGPT | Gemini | Next action |
|---|---|---|---|---|
late payment reminder email | 4 | Cited | Cited | Keep; refresh yearly |
how to ask for overdue payment politely | 9 | Not cited | Cited | Add a direct-answer opening and sample wording |
late payment fees by state | 27 | Not cited | Not cited | Build a dedicated cluster page with a table |
invoice payment terms net 30 | 14 | Not cited | Not cited | Add the missing terms competitors cover |
what to do if a client doesn't pay | 51+ | Not cited | Not cited | New article; it's a fan-out sub-question |
Reading the table: the three keywords neither engine cites all rank 14th or lower, so the work is classic SEO (better coverage, new cluster pages), not rewriting sentences for AI.
What we found: the current top GEO pages
When we ran generative engine optimization through RankTera's KW Research (US Google results, October 15, 2026), the top-ranking pages it analysed (Semrush, Quora, Reply, Coursera and Built In) averaged 1,376 words, and RankTera extracted 62 heading-level keywords. The most repeated were "Generative Engine Optimization", "AI systems" and "Generative Engine Optimization GEO" (3 times per page on average), then "AI visibility", "AI Overviews" and "generative engines" (2 each).
Our SERP analysis the same day found Wikipedia at position 1 and Google's own AI optimization guide at position 3. Across the six top results the median length was 1,953 words, two had an FAQ section and three used at least one table.
What is the best tool for generative engine optimization?
No single tool does GEO for you, because GEO combines SEO, content and measurement. A practical stack is Google Search Console for AI feature traffic, a content research tool to cover the terms and sub-questions the ranking pages cover, and a tracker that records your Google rank and your ChatGPT and Gemini citations for the same keywords. RankTera covers the last two: KW Research and AI Writer for content, and a rank tracker that checks ChatGPT and Gemini citations weekly for every tracked keyword (plans on the pricing page).
Frequently asked questions
What does GEO mean in SEO?
GEO stands for generative engine optimization: optimizing content so AI search engines such as ChatGPT, Gemini, Perplexity and Google AI Overviews retrieve and cite it in their answers. In SEO circles "GEO SEO" means the same thing. Google treats it as part of SEO rather than a separate discipline, because its AI features run on its core ranking systems.
Can a new website get cited by AI search?
Yes. Ahrefs' March 2026 data shows about 62% of Google AI Overview citations came from pages outside the top 10 for the exact query. A new site's best route is ranking for specific long-tail sub-questions with clear answers and original data, not competing head-on for the broad term.
Does llms.txt or schema help GEO?
Not measurably. Google says AI Overviews and AI Mode need no AI text files and no special schema, and that Google Search ignores llms.txt. Ahrefs found 97% of llms.txt files received zero requests, and adding JSON-LD produced no major citation uplift. Keep standard Article markup for hygiene, but put every fact you want cited in the visible text.
How do I get cited by ChatGPT?
Allow OAI-SearchBot in robots.txt, make sure your text is in the server-rendered HTML, and rank for the questions people ask ChatGPT, including the sub-questions. Then check the cited sources for your keywords to see who wins instead. Our ChatGPT SEO guide walks through each step.
Do Google AI Overviews need separate optimization?
Mostly no. Google says a page must be indexed and eligible to show with a snippet, with no extra requirements. The difference is scope: AI Overviews pull from fan-out searches, so covering sub-questions matters more. Our guide to ranking in Google AI Overviews covers the specifics.
Track your GEO results in one table
Generative engine optimization pays off only when you can see which keywords earn citations. RankTera shows your Google position and whether ChatGPT and Gemini cite your domain, keyword by keyword, with the sources they cite instead. Start your 14-day free tier, no card needed.