How to get cited by ChatGPT: OpenAI's index, its bots and the odds

ChatGPT can only cite your page if OpenAI's search crawler, OAI-SearchBot, is allowed to reach it. That is the one eligibility rule OpenAI documents. Everything after it is a matter of odds: whether the question triggers a web search at all, whether your page lands in the pool that search returns, and whether it survives into the written answer. Each of those has public data behind it, and the data is specific to ChatGPT.

Where does ChatGPT search get its results?

From several places at once. ChatGPT search launched in late 2024 leaning on Bing, and two things have changed since.

OpenAI built its own index. Botify's analysis of roughly 7 billion OpenAI log events, reported by Search Engine Journal in April 2026, found OAI-SearchBot activity rose about 3.5× after August 2025, enough to overtake GPTBot. Peec AI's September 2026 analysis of server-side events from 21 May to 21 July 2026 found results tagged with four sources: "Labrador", which Peec identifies as OpenAI's own family of vertical indexes (general web, news, PDF, YouTube, shopping and more), plus three external scraping providers. Across all its data Peec counted at least eight sources, including Microsoft's Web IQ.

Third parties are still in the loop. OpenAI's ChatGPT search help page says ChatGPT search "sometimes partners with other search providers" and typically rewrites your question into one or more targeted queries that it sends to them.

So the accurate description is a blend: OpenAI's own index plus third-party results. Bing visibility is no longer a reliable proxy for ChatGPT visibility (Peec's words: the overlap is "lower than most people think"). Verifying your site in Bing Webmaster Tools is still cheap and still feeds Copilot, but it is not the requirement. The requirement, per that same help page, is to allow OAI-SearchBot and make sure your host or CDN lets traffic from OpenAI's published search-bot IP addresses through.

Which OpenAI bots should you allow?

OpenAI's bot documentation lists four user agents, and each robots.txt setting is independent of the others.

User agent	What it does	If you block it
`OAI-SearchBot`	Surfaces websites in ChatGPT's search features	Your pages are not shown in ChatGPT search answers. OpenAI says robots.txt changes take about 24 hours to apply.
`GPTBot`	Collects content that may be used to train OpenAI's foundation models	Your content is excluded from training. OpenAI documents no effect on search answers.
`ChatGPT-User`	Visits pages for user actions in ChatGPT and custom GPTs, such as opening a page a conversation needs	ChatGPT can't fetch your page live for that user. OpenAI notes robots.txt rules "may not apply" because a person initiated the action.
`OAI-AdsBot`	Checks landing pages submitted as ChatGPT ads	Only matters if you advertise in ChatGPT. Not used for training.

A publisher who wants out of training but into answers needs only this:

User-agent: GPTBot
Disallow: /

User-agent: OAI-SearchBot
Allow: /

The 2024-era "block AI crawlers" templates that disallow everything from OpenAI achieve the training opt-out and remove the site from ChatGPT search at the same time. If you inherited one, this is the first thing to change.

Free tool · No signup

Free AI Crawler Checker

Paste a domain → see which AI crawlers your robots.txt allows and which it blocks, with training and retrieval bots told apart.

Check your crawlers →

How often does ChatGPT cite anything?

Rarely, and more often every month. Similarweb's 2026 Generative AI Landscape report found 6.8% of US desktop ChatGPT answers carried a citation in May 2026, up from about 1.3% in June 2025. That average covers every prompt, including writing and coding tasks with nothing to cite, so it says little about how often a genuine lookup question gets sources. The split by sector says more:

Sector	ChatGPT answers with a citation, May 2026
Travel	22.6%
Retail	13.5%
Sports	10.7%
Finance	8.0%
Education	4.8%

Similarweb via Search Engine Journal. US desktop web only; excludes API, desktop-app and mobile usage.

If your niche sits near the bottom of that table, citations will be sparse whatever you do to the page. Our working assumption (not a measured finding) is that questions which plainly need current or specific information, such as prices, comparisons and recent changes, are the ones most likely to trigger a search. Target those rather than broad definitions.

The audience is large enough to justify the effort: roughly 900 million weekly users as of February 2026 (TechCrunch), and reportedly close to a billion by mid-2026.

Which retrieved pages does ChatGPT actually quote?

About one in seven. AirOps' March 2026 study followed 548,534 pages that ChatGPT retrieved across 15,000 prompts. Only 15% were cited in the final answer.

Three findings from the same data explain who made the cut:

Google rank still predicts it. Pages ranking #1 for the prompt or one of its fan-out queries were cited 43.2% of the time, about 3.5× more often than pages outside the top 20. 55.8% of cited pages ranked in a top 20 somewhere.
Fan-out queries matter. ChatGPT splits a question into several searches, and 32.9% of cited pages were found only through those sub-queries, not the original prompt. A section that answers a likely sub-question can earn the citation on its own.
Position on the page matters. Kevin Indig's Growth Memo study of 1.2 million ChatGPT answers (Search Engine Land, February 2026) found 44.2% of citations came from the first 30% of a page's text, 31.1% from the middle and 24.7% from the final third. Front-loaded, not exclusively so.

Freshness shows up in ChatGPT's choices too. In Seer Interactive's 2026 study of 47,097 citations, 73% of the pages ChatGPT cited had been updated within the past year. Ahrefs found ChatGPT's in-text references averaged 393 days newer than the organic results for the same queries. Change the visible content when you refresh; no engine documents rewarding a new date on unchanged text.

The page-level writing advice (answer first, self-contained sections, attributed claims) is the same for every engine and lives in what is AEO.

Why do ChatGPT citations shift without any change on your side?

Model changes reshuffle them. When OpenAI moved ChatGPT to GPT-5.5 on 22–23 May 2026, SISTRIX saw 47% of citation patterns change overnight in its German dataset, against a normal day-to-day variation of 1–2% (Search Engine Journal). A lost citation in the week of a model release says more about the model than about your page.

That is also why a single check proves nothing. OpenAI offers no publisher dashboard today, so you sample: the same questions, several runs each, spread over days, recorded as a rate. How to track AI citations covers how many runs that takes before the number means anything.

Your first hour on ChatGPT

Open yourdomain.com/robots.txt and confirm nothing (including User-agent: *) disallows OAI-SearchBot.
Check your CDN or WAF bot settings, then confirm OAI-SearchBot hits with 200 responses in your logs.
Write down five questions in your niche that need current or specific information, and run each in ChatGPT with search on. Note which domains get cited and which sentence they quote. That list is your baseline and your competitive brief.

How long after allowing OAI-SearchBot will my site appear in ChatGPT search?

OpenAI says its systems take about 24 hours to pick up a robots.txt change. How long it then takes OAI-SearchBot to crawl your pages, and for them to start appearing in answers, is not documented. Check your server logs for OAI-SearchBot requests before you start measuring citations.

Does the OpenAI API cite the same sources as ChatGPT?

Not necessarily. The API's web search tool returns cited URLs as url_citation annotations, which makes it useful for automated checks, but the model, settings and query rewriting can differ from the consumer app. Treat API results as a related signal and label them separately from what users see on chatgpt.com.

Sources

Updated October 1, 2026: ChatGPT search is now described as OpenAI's own index blended with third-party results; added OAI-AdsBot and ChatGPT-User's live-fetch role; re-attributed the 6.8% figure to Similarweb; removed claims no source supports.