How can a business appear in ChatGPT?
Published on
There is no registration: to appear in ChatGPT is to be found by the searches it runs before answering, then chosen among the pages they return. Three conditions matter in our measurements. Its search crawler, OAI-SearchBot, must be able to read your site. ChatGPT must have a reason to look for you: in 72 to 91% of its searches, it writes the name of a source it already knows. And your pages must rise to the top of its results, on your own site as well as on the third-party sources in your field, because it cites only about one page in ten of what it retrieves.
For what type of question? It all depends on the question your customer asks. If they name your company, ChatGPT looks for you, and your site is the first lever. If they ask a market question without naming you, ChatGPT chooses its own sources, and what decides is whether your name appears on third-party sources. This page covers both cases.
What it means to appear in ChatGPT
On Google, being listed means being in an index that you can check and whose update you can request. ChatGPT does not work that way. When it searches the web, it writes several queries itself, receives a list of pages and cites a few of them. There is no sign-up form and no console to check your presence: a company shows up in ChatGPT at the moment one of its pages rises in those results and is chosen.
In the app, ChatGPT can also answer without searching, from what it learned during training. A company that had little presence on the web before the model’s knowledge cutoff date is not part of it, and there is no way to get in afterwards. Our measurements cover answers with search, the ones where a page published today can count.
First condition: let OpenAI’s search crawler in
OpenAI states it in its documentation: to appear in ChatGPT’s search answers, a site must let its OAI-SearchBot crawler through, and a site that blocks it is not shown there. This setting is made in the robots.txt file. It is independent of the one for GPTBot, the crawler that collects training data: a site can refuse training and stay open to search.
Our measurements agree. Across a dozen business sectors queried in French, fewer than 1% of the sites retrieved by ChatGPT’s search blocked OAI-SearchBot, and none of the sites it cited; 3% of the cited sites did, however, block GPTBot. You also need to check that a firewall or a bot-protection service does not block OAI-SearchBot before the file is even read. This check removes one cause of total absence; it gets no one cited.
Second condition: ChatGPT needs a reason to look for you
ChatGPT does not search with your customer’s words. It rephrases the question into ten or so searches at the median, and in 72 to 91% of them depending on our corpus, it writes the name of a specific source: a publisher, a government body, a media outlet. It then cites that named publisher 1.8 to 2.6 times more often. On our questions asked in French on September 15, 2026, 95% of its answers also included at least one search in English.
When the question names your company, this reflex works for you: on troubleshooting questions about business software, the documentation of the brand asked about makes up 58% of ChatGPT’s citations. When it does not name you, it works against a company it does not know: it writes the names of the sources it already has in mind, and a company absent from those sources is neither searched for nor found. On a market question, appearing in ChatGPT therefore starts with being mentioned somewhere other than on your own site.
Third condition: be present in the sources it cites
On questions that name no one, ChatGPT mostly cites third-party sources. On our troubleshooting questions asked without naming a brand, government bodies accounted for 45% of its citations. Muck Rack, across more than 25 million links cited by ChatGPT, Claude, and Gemini in seventeen sectors, counts in May 2026 13.7% of citations going to the content of the company targeted by the question; the rest went to the press, encyclopedias, institutions, or other companies.
Not all platforms are equal. With ChatGPT, the Reddit, YouTube, and Wikipedia pages retrieved by its search were cited between 0 and 1.5% of the time depending on the corpus, compared with 10 to 15% for other pages. Being present on these platforms brings a name into what ChatGPT reads, not into what it displays. The sources that count are those that cover the questions of your market with pages that answer them: specialized media, reference sites, guides in your sector.
Fourth condition: pages that rise and get cited
Even when found, a page still has to be chosen. ChatGPT cites about one page in ten of what its search returns, and rank carries heavy weight: a page in the top three positions of its results is cited 6.9 to 8.5 times more often than a page beyond the ninth, across all our corpora. It also more often discards thin pages, under 300 words, and slow pages.
The form of the page counts next. With ChatGPT, a page whose title repeats the question asked is cited 1.3 to 1.9 times more often depending on the corpus, and a separator in the title, such as “Question | Site”, 1.5 to 1.8 times more. A good Google ranking is not enough: according to Ahrefs in August 2025, only 8% of the pages cited by ChatGPT were in Google’s top 10 for the same query. The rank that counts is the one in the searches ChatGPT writes itself.
What has no proven effect
Several recipes for showing up in ChatGPT circulate without standing up to measurement. The llms.txt file: no major AI publisher documents using it to choose sources, and among the sites retrieved by the search of ChatGPT, Claude, and Gemini, the cited sites do not publish one significantly more often than the others. Schema.org markup: Ahrefs measured no clear rise in citations after adding it in 2026, and Google writes that no particular markup is needed for its own AI answers.
A single test proves nothing either, one way or the other. When we asked the same questions again about an hour later, only 25% at the median of the pages ChatGPT cited the first time were cited again. Presence is measured as a frequency, over many questions asked several times, never on a single answer.
What this changes for a business
The order of work follows from these conditions. First what can be checked in a few minutes: OAI-SearchBot’s access to your site. Then pages that clearly answer the questions your customers ask, in the languages ChatGPT searches in, with the question in the title. Finally, for the questions that do not name you, a regular presence on the third-party sources in your field, so that ChatGPT has your name in mind when it writes its searches.
This last point takes the longest, and it is the one FaireDuBruit handles: every week, one article on your site and three on our specialist media, in your language and in English. Our media publish articles written from your facts, which are not independent opinions. We measure AI reads on these media, not yet the effect of these publications on their citations, and no citation is guaranteed.
What we do not know
Our measurements cover GPT-5 through its API, with web search enabled: the ChatGPT app may behave differently, notably by answering without search. They describe what ChatGPT searches for, retrieves, and cites; they do not measure the effect, for a company, of opening its site, of a publication, or of a mention.
Sources
- OpenAI — OpenAI crawlers, .
- Muck Rack — “What is AI Reading?” (Generative Pulse), p. 5, .
- Ahrefs, Louise Linehan — overlap between AI search and top 10, .
- Ahrefs, Louise Linehan — adding structured data and AI citations, .
- Google Search Central — “Optimizing your website for generative AI features on Google Search”, .
- FaireDuBruit measurements, from September 4 to 16, 2026 — see “How we measured”.