GEO: how to get cited by ChatGPT and AI search.

Mostly good SEO with sharper edges: get into every index the assistants search, answer plainly, earn real mentions and measure honestly.

A thick stack of folded newspapers
Photo by Tim Mossholder on Unsplashdithered by Cyborb

To get your site cited by ChatGPT and other AI search tools, first make sure each assistant’s search can reach your pages: allow the right crawlers and get into the indexes they use. Then make each page the clearest answer to a specific question, and earn mentions on the sites assistants already trust. That is generative engine optimization in 2026, and it is mostly good SEO with sharper edges.

Some popular advice has not held up. The rewriting tactics from the 2024 study that named GEO failed later tests on realistic systems, and Google says its AI features need no special files or markup. Here is how assistants choose sources, what the evidence supports now, how to track your citations, and one Cloudflare setting to check today.

The short version
  • Assistants cite what their search finds. Allow OAI-SearchBot, PerplexityBot and Claude-SearchBot, stay indexed in Google, Bing and Brave, and make sure Cloudflare is not blocking Googlebot.
  • Blocking a training crawler such as GPTBot keeps you in search answers, but future models may know less about you. Google-Extended also covers Gemini app answers.
  • The 2024 GEO study’s 30 to 40% gains did not hold up in later tests. Ranking in ordinary search, clear answers and real mentions matter more.
  • Tricks backfire: Google’s spam policy now covers manipulating AI answers, and self-ranking “best X” lists often get rivals recommended instead.
  • Track AI impressions in Search Console, citations in Bing and utm_source=chatgpt.com referrals, and repeat prompt checks without forcing search.

How AI assistants choose which sources to cite

AI search works in two steps. First the assistant searches; Google’s AI features run several related searches at once, a technique Google calls query fan-out. Then the assistant reads the best passages and writes an answer that links to its sources. If your page is not in what the search step returns, nothing on the page matters.

At Google, that search step is ordinary Search: a page must be indexed and able to show a snippet to be eligible. Since August 31, 2026, the site must also be set to Include in a new Search Console setting, which is the default.

What gets picked also leans toward reference sources. In Pew’s March 2025 data, Wikipedia, YouTube and Reddit made up 15% of the sources in Google’s AI summaries, and government sites appeared three times as often as in regular results.

Which crawlers and indexes to let in

Each company runs its own crawlers, and most separate search from model training. Some also rely on another company’s index:

AssistantAllow this to appear in answersTraining is controlled by
ChatGPT searchOAI-SearchBotGPTBot
PerplexityPerplexityBotNo training bot: Perplexity says PerplexityBot does not crawl for foundation models
ClaudeClaude-SearchBot, plus Googlebot: Claude’s search uses Brave, whose crawler follows your Googlebot rulesClaudeBot
Microsoft CopilotBingbot, because Copilot runs on Bing’s indexBing’s guidelines list no training token
Google AI Overviews and AI ModeGooglebot, with Search Console’s generative AI setting on IncludeGoogle-Extended, which does not affect Search
Gemini appGooglebot, and not blocking Google-Extended, which also controls groundingGoogle-Extended

OpenAI is explicit: sites that opt out of OAI-SearchBot “will not be shown in ChatGPT search answers”, though they can still appear as plain navigational links. Blocking GPTBot, by contrast, only signals that your content should not be used for training. For every crawler and what blocking it does, see our AI crawlers list.

Two rows need care. First, Anthropic lists Brave Search as a web search provider for Claude, and Brave’s crawler has no user agent of its own: it crawls only what your robots.txt lets Googlebot crawl. So allowing Claude-SearchBot alone is not enough.

Second, Google-Extended is not a pure training switch. It also decides whether the Gemini app and Vertex AI may use your pages to ground answers. Google says it does not affect Google Search, so AI Overviews and AI Mode follow the Search Console setting instead.

What the evidence says about generative engine optimization

The study that named GEO rewrote web pages with different tactics and measured how much of each AI answer drew on them. Citing sources, adding quotations and adding statistics worked best, raising visibility by 30 to 40%, while keyword stuffing did little.

But that was a 2023 setup: a simulated engine built on GPT-3.5, plus a test on Perplexity. Later studies tried the same ideas in more realistic conditions, and the tactics mostly failed.

30 to 40%
Visibility gain from citing sources, quotes and statistics on a 2023 research engine
GEO study, KDD 2024
2 of 2
Peer-reviewed benchmarks since then that found those tactics largely ineffective in realistic setups
C-SEO Bench (NeurIPS 2025), SAGEO Arena (KDD 2026)
120%
More clicks per impression for brands cited in a Google AI Overview, on informational queries
Seer Interactive, April 2026
  • C-SEO Bench (NeurIPS 2025): most such methods were “largely ineffective” and often harmful, while traditional SEO was “significantly more effective”. Gains shrank as more sites used the same trick.

  • SAGEO Arena (KDD 2026): tested over real pages through retrieval, reranking and writing, existing methods were “largely impractical under realistic conditions”.

  • A September 2026 preprint, not yet peer reviewed: re-measured on ten modern engine families, the 2023 tactics moved citations on none.

So cite sources and give dated numbers because readers trust them, not to trick a model. The lever that held up is plainer: rank well where the assistant searches.

Google has closed the door on tricks, too. Its spam policy, updated August 28, 2026, covers “attempting to manipulate generative AI responses in Google Search”. Its May 2026 guide adds that separate pages for each variation of a query, made to manipulate AI answers, break its scaled content abuse policy.

For more numbers, see AI search statistics for 2026. For how AI answers changed clicks overall, see SEO in the age of AI.

Backed by evidence or official docs
  • Letting search crawlers in and staying indexed (vendor docs)
  • Ranking well in ordinary search (C-SEO Bench)
  • One strong page per topic instead of near-duplicates (Bing)
  • Accurate dates that match your sitemap (Google, Bing)
  • Mentions on YouTube and across the web (Ahrefs, correlation only)
Not proven, not needed or risky
  • Rewriting pages with quotes and statistics as an AI trick (did not hold up)
  • Keyword stuffing (little to no effect even in the GEO study)
  • Schema as a citation lever (no lift in Ahrefs’ test)
  • llms.txt as a citation lever (Google Search does not use it)
  • Self-ranking “best X” lists (often help rivals instead)
  • Mass-produced pages, including one per query variation (Google treats them as spam)

How to make a page worth citing

  1. Let the search crawlers in

    Check your robots.txt. Blocking a search crawler removes you from that assistant’s answers. The last rule below is optional: it keeps your pages out of OpenAI’s model training without affecting ChatGPT search.

    robots.txt
    User-agent: OAI-SearchBot
    Allow: /
    
    User-agent: PerplexityBot
    Allow: /
    
    User-agent: Claude-SearchBot
    Allow: /
    
    User-agent: GPTBot
    Disallow: /

    Weigh the GPTBot rule first. Language models answer best about things they saw in many documents during training, so blocking training crawlers makes future models less likely to know your brand at all.

    Two more checks. A crawler with its own group ignores the * group, so copy any private Disallow lines into each named group. And on Cloudflare, check the setting in the warning above.

  2. Get into every index assistants search

    Submit your sitemap in Google Search Console and Bing Webmaster Tools. In Search Console, open the Search generative AI setting and check that it is set to Include, the default. For Claude, check that Brave Search shows your key pages; Brave’s submit page lets you ask it to re-fetch one.

  3. Answer the question in the first two sentences

    Lead with the answer in plain words, then explain. Bing’s team notes that a direct question with a clear answer mirrors how people search, and that assistants “can often lift these pairs word for word” into their answers.

  4. Back claims with sources, numbers and quotes

    Link to the sources you rely on, give specific numbers with dates, and quote named experts where it helps. Bing asks for “examples, data, and cited sources”. Do it for readers: as an AI trick, it did not hold up.

  5. Keep one strong page per question

    Merge near-duplicate pages and point the rest at the survivor with a canonical tag. Bing warns that AI systems group near-duplicate URLs and may pick an outdated version to represent the group.

    When you merge or remove pages, tell search engines through IndexNow, a protocol for announcing changes. Bing says duplicate and low-value URLs slow how fast new information reaches AI summaries. Google does not take part in IndexNow.

  6. Keep dates honest

    Show a labeled Published or Last updated date, and keep it identical to datePublished and dateModified in your structured data, as Google asks. Bing says freshness signals directly affect how quickly updates reach AI answers.

    But change dates only when the content changes. Google uses your sitemap’s lastmod only if it is “consistently and verifiably” accurate, and Bing says not to set it to the time the sitemap was generated.

  7. Earn mentions beyond your own site

    Brands that people talk about show up more in AI answers. In Ahrefs’ study of 75,000 brands, YouTube and web mentions correlated with AI visibility far more than backlinks did. Original data, useful tools and clear opinions are what get you mentioned.

    Skip the shortcuts. Google says seeking inauthentic mentions “isn’t as helpful as it might seem”. And in Lily Ray’s check of 100 “best X” queries, AI Overviews cited a company’s self-ranking list but left that company out of the recommendation in 224 of 323 cases.

What about schema markup and llms.txt?

Structured data, or schema, helps search engines understand a page and can earn rich results. For AI citations the evidence is mixed. Google’s May 2026 guide says schema “isn’t required for generative AI search”, while Bing says it can label content “as a product, review, FAQ, or event”.

In Ahrefs’ before-and-after test of 1,885 pages that added schema, against 4,000 that did not, citations rose 2.2% in ChatGPT and 2.4% in AI Mode, both statistically zero, and fell 4.6% in AI Overviews. But every page in it was already cited, so it cannot say whether schema helps a page get found in the first place.

A KDD 2026 lab benchmark, by contrast, found that structural information such as schema helped in its test search pipeline. Use schema where it earns rich results, and do not expect it to win citations on its own.

llms.txt is a proposal, not a standard. Jeremy Howard suggested it in September 2024: a Markdown file at /llms.txt that gives language models a short summary of a site with links to key pages. A second version followed in August 2026.

Its real use is documentation. OpenAI, Anthropic and Google publish one for their developer docs, and Chrome’s Lighthouse checks for the file, but the check is unscored and a missing file counts as not applicable.

For AI search, Google’s guide says Google Search does not use AI text files, though keeping one for other systems is “completely fine”. OpenAI’s publisher FAQ does not mention llms.txt. Add one if developers read your docs through coding agents, but do not count on it for citations.

How to track whether AI search cites you

  • Google Search Console. The Generative AI performance report, live for all sites since August 31, 2026, counts impressions in AI Overviews and AI Mode by page, country, device and date, once you have enough. It shows no clicks; those stay mixed into the “Web” search type.

  • Bing Webmaster Tools. The AI Performance report, launched in public preview on February 10, 2026, shows how often Copilot, Bing’s AI summaries and some partner integrations cite your pages, which pages, and the “grounding queries” behind them. It shows no clicks either.

  • Analytics referrals. ChatGPT adds utm_source=chatgpt.com to the links it sends, so those visits are easy to separate in tools such as Google Analytics.

  • Server logs. Crawler visits show access, not citations, but a page no search bot has fetched cannot be cited.

Terminal
# Count visits from AI search crawlers and user-triggered fetchers
grep -Eio "OAI-SearchBot|ChatGPT-User|PerplexityBot|Perplexity-User|Claude-SearchBot|Claude-User" access.log | sort | uniq -c

Finally, check by hand each month. Ask the questions your pages answer the way a customer would, without telling the assistant to search, and run each one several times. Log whether it searched, which URLs it cited and whether it named you.

Expect noise: answers vary between runs, so read repeated checks as a range, as a 2026 paper on measuring AI visibility recommends. Forcing search also changes the answer: in Graphite’s test of 100 comparison prompts in ChatGPT, it shifted brand visibility by 20 points on average for prompts that never search on their own.

Keep a forced-search prompt like this one as a separate diagnostic: it shows whether an assistant can find your page at all, not what real users see.

PromptFindability check (forces search, so log it separately)
I am researching [topic]. Answer this question: [the exact question your page answers].
Search the web and cite every source you use, with its full URL.

For setup steps in each tool, see how to track AI traffic. For sharper test prompts, see our prompt engineering guide. If you rely on these tools for your own research, our guide to AI research tools shows how to check every citation they give you.

FAQ

What is generative engine optimization?

It is the practice of shaping content so AI systems that write answers are more likely to use and cite it. The name comes from a 2023 research paper that measured which rewriting tactics raised a page’s visibility in AI answers. Today it mostly means being easy for each assistant’s search to find, and being the clearest answer when it does.

Is GEO different from SEO?

Mostly not. Google says “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” Assistants search before they answer, so crawl access, indexing and quality still decide whether you are in the running. GEO adds a sharper focus on being quotable and being mentioned elsewhere.

Does llms.txt help me get cited by ChatGPT?

Not according to the published guidance. OpenAI’s publisher FAQ asks you to allow OAI-SearchBot and does not mention llms.txt, and Google says Google Search does not use AI text files.

Should I block AI crawlers?

Block training crawlers if you do not want your content used to train models, knowing future models may then know less about you. OpenAI says blocking GPTBot only signals that content should not be used for training, and Anthropic uses a separate crawler for search. At Google, blocking Google-Extended also stops the Gemini app from using your pages to ground answers. Blocking a search crawler such as OAI-SearchBot takes you out of that assistant’s answers.

How do I leave AI Overviews but stay in Google Search?

Use the Search generative AI setting in Search Console, live for all sites since August 31, 2026. Choosing Exclude removes your links and content from AI Overviews, AI Mode and Discover’s AI features, usually within a few days, without affecting the rest of Search. It does not stop Gemini training: that is Google-Extended.

Key takeaways
  • Get found first: let each assistant’s search crawler in, stay indexed in Google, Bing and Brave, and check Cloudflare’s AI settings.
  • Make pages worth quoting: answer first, cite sources and add dated numbers, for readers rather than as a trick.
  • Keep one strong page per question, keep dates honest, and tell search engines when pages change.
  • Earn real mentions elsewhere; skip self-ranking lists and pages for every query variation.
  • Track AI impressions in Search Console, citations in Bing, referrals in analytics, and unforced answers by hand.

Next, read SEO in the age of AI, look up any bot in our AI crawlers list, or build a website with AI that follows these rules from day one.

Sources
  1. GEO: Generative Engine Optimization, Aggarwal et al., KDD 2024
  2. C-SEO Bench: Does conversational SEO work?, Puerto et al., NeurIPS 2025
  3. SAGEO Arena: A realistic environment for evaluating search-augmented generative engine optimization, Kim et al., KDD 2026
  4. Scoring without the engine, Bajemon and Rochet, arXiv preprint, September 2026
  5. Don’t measure once: measuring visibility in AI search, Schulte et al., arXiv, April 2026
  6. Do not force search in prompt tracking, Graphite, August 2026
  7. Large language models struggle to learn long-tail knowledge, Kandpal et al., ICML 2023
  8. Optimizing your website for generative AI features on Google Search, Google Search Central, May 2026, updated July 2026
  9. Spam policies for Google web search, Google Search Central, August 2026
  10. AI features and your website, Google Search Central, December 2025
  11. Search generative AI control, Search Console Help, August 2026
  12. Generative AI performance report, Search Console Help, August 2026
  13. Google’s common crawlers, Google, July 2026
  14. How Google interprets the robots.txt specification, Google, August 2026
  15. Influence your byline dates in Google Search, Google Search Central, December 2025
  16. Build and submit a sitemap, Google Search Central, July 2026
  17. Overview of OpenAI crawlers, OpenAI
  18. Publishers and developers FAQ, OpenAI Help Center, August 2026
  19. Perplexity crawlers, Perplexity
  20. Does Anthropic crawl data from the web?, Anthropic, April 2026
  21. Subprocessors, Anthropic Trust Center
  22. Brave Search crawler, Brave
  23. Bing Webmaster Guidelines, Microsoft Bing
  24. Optimizing your content for inclusion in AI search answers, Microsoft Bing, October 2025
  25. Does duplicate content hurt SEO and AI search visibility?, Microsoft Bing, December 2025
  26. Keeping content discoverable with sitemaps in AI powered search, Microsoft Bing, July 2025
  27. Introducing AI Performance in Bing Webmaster Tools public preview, Microsoft Bing, February 2026
  28. IndexNow FAQ, IndexNow
  29. Your site, your rules: new AI traffic options for all customers, Cloudflare, July 2026
  30. Block AI bots, Cloudflare Docs, July 2026
  31. Cloudflare just changed how AI crawlers scrape the internet-at-large, Cloudflare, July 2025
  32. Google users are less likely to click on links when an AI summary appears in the results, Pew Research Center, July 2025
  33. AIO impact on Google CTR: 2026 update, Seer Interactive, April 2026
  34. Top brand visibility factors in ChatGPT, AI Mode and AI Overviews, Ahrefs, December 2025
  35. We tracked 1,885 pages adding schema. AI citations barely moved., Ahrefs, May 2026
  36. Why calling yourself the “best” could be helping your competitors win in AI search, Lily Ray, June 2026
  37. The llms.txt proposal, llmstxt.org, August 2026
  38. llms.txt audit, Chrome for Developers, May 2026
cyborb.ai

Stop reading about it. Build it.

Describe what you want in plain words. Cyborb plans the work, writes and runs the code, makes the assets, and puts the result online.

Download Cyborb

Free to start. No card required.