How to Get Cited by ChatGPT

Getting cited by ChatGPT starts with a permission most advice skips. OpenAI says ChatGPT search answers leave out sites that refuse its search crawler, and several of the outlets brands chase refuse it. After that, the work is the familiar part done precisely: coverage in sources ChatGPT already uses for your buyers' questions, carrying one claim specific enough to cite.

The order matters. A placement in a publication that blocks the crawler can be excellent PR and still be invisible in ChatGPT.

ChatGPT search reads only what OAI-SearchBot may read

OpenAI documents the rule itself. Its crawler overview says OAI-SearchBot is used to surface websites in ChatGPT's search features, and that sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links. GPTBot is a separate setting that covers model training. ChatGPT-User, the agent that fetches a page when a person asks, is not used to decide what can appear in search.

So there are two switches, and they do different jobs. A publisher can refuse training and still be eligible for ChatGPT search, or refuse search and lose the citation slot entirely.

What the Index shows at the head of the market

The Machine Relations Index asks six AI engines the same buying and research questions every day and records every source they cite. In release mri_score_v2.1 of 2026-10-14, covering 2026-05-10 to 2026-10-14, 89 of the 100 most-cited domains drew at least one ChatGPT citation.

Of the 11 that drew none, 7 served a robots.txt on 2026-10-14 that disallows OAI-SearchBot across the whole site. Four name it directly: The New York Times, PCMag, CNET and Everyday Health. Three reach it through a blanket rule for every crawler they do not name: Reddit, Facebook and Instagram. All seven were cited by Perplexity and Google AI Mode in the same window.

Reddit is the sharpest case. It is the most-cited domain in the entire Index, cited in 2,429 of 25,323 answer runs, and its Index page lists four engines that cited it. ChatGPT is not one of them.

It is an association, not a switch. Amazon, Gartner and CNBC also disallow OAI-SearchBot by name today and still drew ChatGPT citations in the window. Investopedia allows the crawler and drew none. Robots files change, and a site's rules on one day say nothing certain about the five months before it. The training-token versus search-token split across every engine is in the Machine Relations study of robots.txt disallows.

The operating rule is simple. Before a placement goes on a ChatGPT plan, read the outlet's robots.txt for an OAI-SearchBot group and for a blanket disallow. Then read your own. A rule written to keep content out of model training is harmless here only if it names GPTBot, not the search crawler.

ChatGPT's source list is its own

The second finding matters more for how you measure. Across the window, ChatGPT cited 8,154 distinct domains, and 4,727 of them, 58%, were cited by none of the other five engines. The same figure is 49% for Perplexity, 37% for Gemini and 29% for Claude.

That makes ChatGPT the engine whose sources overlap least with the rest. A blended "AI visibility" number averages it away. If your buyers use ChatGPT, measure ChatGPT on its own, and check which sources it actually cites for your questions before you decide where to pitch.

ChatGPT does at least tend to look. In the same release, 5,026 of 5,704 ChatGPT runs returned a citation, against 2,090 of 5,803 for Gemini, which more often answers from memory. The Index reaches ChatGPT through OpenAI's API model, gpt-4.1-mini, so the consumer app may search more or less often; read it as what the instrument observed, not a constant of the product.

Then the work that compounds

Once the crawler can read the source, ChatGPT chooses among sources the way every engine does: by what answers the question. This is the part of Machine Relations, the discipline I named in 2024, that every engine shares. Earned media is evidence a machine can retrieve, and it only counts when it says something specific.

1. Earn coverage in outlets ChatGPT can read and already cites. The Index's domain pages show which of the six engines cited a source in the window. An outlet that blocks OAI-SearchBot, or that ChatGPT has never cited, is a weak ChatGPT bet no matter how prestigious it is.
2. Make the coverage carry one claim a model can lift. A number with its window, a named customer outcome, a position someone could quote. A profile that calls you innovative gives ChatGPT nothing to cite.
3. State your own facts where ChatGPT will find them. Your site remains the first-party source for what you sell, what it costs and who runs it, and it needs the same crawler access you checked on the outlet. Independent coverage corroborates; it does not replace you.
4. Re-run the same questions on a schedule. Keep a saved baseline of questions, answers and citations, and compare like with like. A citation, a mention, a recommendation and a referral are four separate outcomes; record each without pretending one proves the others.

What I would ignore

OpenAI publishes no citation score, no ranking formula for search sources and no count of placements that earns a citation. When a checklist supplies one, it is describing someone's sample, not OpenAI's system. Schema, headings and freshness are good hygiene, and none of them overrides the crawler permission or a source with nothing to say.

The order is the whole method: can ChatGPT read the source, does ChatGPT already use sources like it for your questions, and does the source say something worth citing. The same mechanism connects PR to AI search visibility on every engine; Gemini adds its own permission layer, covered in how to get cited in Gemini. AuthorityTech's ChatGPT guide develops the execution.

Method note. Engine lists, citation counts and run totals come from the Machine Relations Index public release and its release manifest, release mri_score_v2.1+2026-10-14+7a624b28cd41. "Most-cited" means the top 100 ranked domains in that release. Robots.txt files were fetched on 2026-10-14; one of the 11 domains without a ChatGPT citation refused the request and is counted as unread.