When you ask an assistant “which osteopath is open on Saturdays in Rennes?” or “gluten-free bakery near the station”, it no longer answers from memory alone: it runs searches, reads pages and cites the ones that helped it. In France, 48% of people aged 12 and over used generative AI in 2025, against 20% in 2023, and 73% of these users use it to look for information. ChatGPT leads by far, ahead of Gemini (Crédoc for Arcep, 2026).
So what these assistants say about you depends on what they find. Here is how each one reads websites, according to its own documentation, and what really helps you show up there.
Google: the usual rules are enough
Google’s AI Overviews and AI Mode rely on its search engine’s index. Google writes it plainly: there are no additional requirements to appear there, and no special optimization. A page can serve as a supporting link as soon as it is indexed and can be shown with a snippet in the classic results. Google also warns that nothing is guaranteed, and that AI Overviews only show when it deems them useful.
To answer, these features run several searches close to the question asked, which lets them show more varied links than an ordinary results page. So a precise page, answering a precise question, stands a chance against big sites.
Two practical details:
- the directives that limit snippets (nosnippet, max-snippet) or indexing (noindex) apply to these features too: a page denied a snippet can’t serve as a supporting link;
- visits coming from these answers are counted in Search Console with the rest, under the “Web” search type.
Gemini: a token worth knowing, Google-Extended
The Gemini app can rely on Google’s index to back up its answers. For this, Google provides a name to use in robots.txt, Google-Extended. It decides whether your pages can be used to train future Gemini models, and also to ground the Gemini app’s answers. It changes nothing about your presence in Google Search, nor your ranking.
So refusing Google-Extended doesn’t remove you from AI Overviews, which go through Googlebot. But your pages may then stay out of the Gemini app’s answers.
ChatGPT: let OAI-SearchBot through
OpenAI documents several bots, each with its role:
- OAI-SearchBot crawls the web for ChatGPT search. A site that blocks it no longer appears in search answers, except as a mere navigation link.
- GPTBot collects pages that may be used to train OpenAI’s models. Blocking it means refusing that use.
- ChatGPT-User visits a page when a user asks it to. OpenAI specifies that robots.txt may not apply to these visits, since a person triggered them.
These settings are independent: you can refuse training and stay visible in search. OpenAI counts about 24 hours for a robots.txt change to be taken into account by its search.
Perplexity and Claude
Perplexity documents two bots. PerplexityBot is used to surface and link sites in its results; Perplexity specifies that it trains no model, and advises letting it through to appear. Perplexity-User fetches a page at a user’s request and, by Perplexity’s own admission, generally ignores robots.txt. There too, allow up to 24 hours after a change.
Anthropic, the company behind Claude, documents three: ClaudeBot for training, Claude-SearchBot for the quality of search answers, and Claude-User for visits requested by a user. Anthropic says its bots respect robots.txt, and warns that blocking Claude-SearchBot or Claude-User can reduce your visibility.
| Service | Search | Training | Visit requested by a user |
|---|---|---|---|
| Googlebot | Google-Extended (a name, with no bot of its own) | not documented here | |
| ChatGPT | OAI-SearchBot | GPTBot | ChatGPT-User |
| Perplexity | PerplexityBot | none announced | Perplexity-User |
| Claude | Claude-SearchBot | ClaudeBot | Claude-User |
What really helps
None of these services asks for a special file or markup reserved for AI. Google says so explicitly, structured data included. What counts is what every good website already does:
- Be indexed. Check in Search Console that Google knows your page, and that your robots.txt lets the search bots above through. Google advises also checking the settings of your host or CDN, which can refuse bots without robots.txt having anything to do with it.
- Write the facts plainly, as text. Your opening hours, your address, your prices, your service area, what you do and what you don’t. “Open Tuesday to Saturday, 7 a.m. to 7:30 p.m.” reads effortlessly; opening hours on a photo of the shop window, much less so. Google recommends keeping the bulk of the content as text.
- Keep these facts up to date, the same everywhere. An assistant that reads summer hours in November will repeat them. Update the website, the business profile and social media on the same day: Google in fact counts an up-to-date business profile among its good practices.
- Answer the questions people ask you. The ones from the phone (“do you take cards?”, “is there parking?”) are also the ones people ask assistants. A questions-and-answers section offers them clear answers to cite.
- Keep honest structured data, if you publish any: Google asks that it match the visible text of the page.
Checking what they say about you
- Ask them the question, as a customer would: “opening hours of [your name] in [your town]”, “good [your trade] in [your neighborhood]”. Read the answer, and above all the sources cited.
- Fix it at the source. If a piece of information is wrong, it comes from a page: your website, your Google profile, a directory that got it wrong. That is what needs fixing.
- Open your robots.txt, at your-domain.com/robots.txt, and check that no search bot is refused there.
- Don’t count on readers being vigilant: only 21% of generative AI users systematically check what it tells them (Crédoc for Arcep, 2026). What the assistant says about you is often what the customer remembers.
In short: to get cited by AI, be a good search result first. Let their search bots through, write your facts plainly and keep them up to date. No special file or markup is required.