Getting cited by ChatGPT, Perplexity, Claude or Gemini is not obtained through a form or a trick: an assistant cites a page it could read, whose content it understood, whose author it can identify, and which answers the question asked. This page follows those steps in order, with what each vendor documents, the GEO test check to read, and the page that details each fix. AI visibility starts there, not with a ranking.
1. What decides a citation, according to the vendors
What the vendor documents. OpenAI documents three crawlers and their effect. OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features", and "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers". ChatGPT-User is used "for certain user actions in ChatGPT and Custom GPTs" and "is not used for crawling the web in an automatic fashion". GPTBot "is used to crawl content that may be used in training" the models. Nothing in that documentation describes how a source is chosen among those the crawler can read (OpenAI, bots and crawlers).
What we recommend. Start from what is documented and stay there: a citation first requires that the answer crawler gets in, then that it finds text, then that it knows who is speaking. These are prerequisites, not guarantees, and each is checked on your home page by the GEO test, which reports AI answer crawlers allowed, text readable without JavaScript and the identity entity. Everything said beyond that, about how an assistant ranks its sources, is interpretation; we flag it as such when we offer one.
2. Being read: robots.txt and CDN
What the vendor documents. Anthropic describes Claude-SearchBot, which "navigates the web to improve search result quality for users", and Claude-User, which "may access websites" when "individuals ask questions to Claude". Its bots "respect 'do not crawl' signals by honoring industry standard directives in robots.txt", are refused through a dedicated group, and their IP addresses are published (Anthropic, web crawling).
What we recommend. One robots.txt group where every answer crawler is named and shares the rules of *, and a check from outside with each crawler's user agent, because a CDN or a firewall can return a challenge the file does not announce. The file, crawler by crawler, is in our French guide Crawlers IA et robots.txt; the proxy case in Cloudflare and AI crawlers.
How to fix
User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: Claude-SearchBot
User-agent: Claude-User
User-agent: PerplexityBot
User-agent: *
Allow: /
Disallow: /admin/3. Being readable: the text in the HTML
What the vendor documents. Google is the only vendor to document JavaScript execution by its crawlers, with a rendering queue where a page "can take longer than" a few seconds, and it recommends server-side rendering because "not all bots can run JavaScript" (Google, JavaScript SEO basics). OpenAI, Anthropic and Perplexity document nothing on this point.
What we recommend. Assume an assistant's fetcher reads the raw HTML: title, headings, paragraph text, prices and structured data must be in the HTTP response, without a script. A direct answer at the top of the page, in the first lines, is what an assistant can pick up; content that arrives after an AJAX call does not exist for it. Server rendering, stack by stack, is in our French guide le contenu sans JavaScript.
4. Being identified: Organization, sameAs, llms.txt
What the vendor documents. Organization structured data "can help Google better understand your organization's administrative details and disambiguate your organization in search results": identifiers such as vatID, iso6523Code, taxID or duns, and sameAs links to pages about the organization elsewhere (Google, Organization structured data). No assistant vendor documents its use of this data; none documents the llms.txt file.
What we recommend. An Organization node on every page, with a registry identifier, address, founder and exact profiles, and a sentence that rules out namesakes; an llms.txt that repeats that identity as text. An assistant hesitating between two entities of the same name cites neither. The graph is in Schema Organization et identité d'entité (French), the file in llms.txt (French).
How to fix
{
"@type": "Organization",
"name": "Your brand",
"legalName": "YOUR COMPANY LTD",
"identifier": {"@type": "PropertyValue", "propertyID": "Company number", "value": "12345678"},
"sameAs": ["https://www.linkedin.com/company/your-brand"],
"description": "Services agency in Bougival, France. Not to be confused with the software of the same name."
}5. Assistant by assistant: ChatGPT, Perplexity, Claude, Gemini
What the vendor documents. Perplexity distinguishes PerplexityBot, "designed to surface and link websites in search results on Perplexity", which is "not used to crawl content for AI foundation models" and respects robots.txt, from Perplexity-User, which "generally ignores robots.txt rules" because a user requested the fetch (Perplexity, crawlers). ChatGPT and Claude follow the same pattern, a search crawler and a user fetcher, documented in sections 1 and 2.
What we recommend. The same site serves all four. ChatGPT: OAI-SearchBot allowed, and a page that answers in its first lines. Perplexity: PerplexityBot allowed, and dated pages, because its answers cite sources with their date. Claude: Claude-SearchBot and Claude-User allowed, without forgetting the CDN. Gemini and Google's AI Mode: Google-Extended has no effect on Search according to Google, and AI Mode has "no additional requirements"; what ranks on Google can appear there. None of the four offers URL submission or an equivalent of Search Console.
6. Authority and reputation: what is not documented
What the vendor documents. Google states that "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary", and that sites appearing in AI features "are included in the overall search traffic in Search Console" (Google, AI features in Search). No assistant vendor documents the weight of third-party mentions, customer reviews or a Google Business Profile in its answers.
What we recommend. This section is our opinion, and nothing in it is documented by a vendor. We observe that the pages cited are often the ones other sites already cite, and that customer reviews and directory listings come back in answers about local businesses. We draw two cautious pieces of advice: keep your official profiles consistent with your site, same name, same address, same activity; and earn mentions on sites your clients read. We promise no effect and refuse to put a number on one.
What we did on seoforge.fr
The four prerequisites are implemented on this site and verifiable: a robots.txt naming every answer crawler, server-rendered pages, an Organization node with the company registration number and a clause against namesakes, an llms.txt generated from the site map. We publish no citation rate: we have not measured one in a publishable way, and the GEO test does not measure it.
Going further
This page follows the plan of the AI search optimisation guide, which details each dimension with its sources. If you have applied everything and are still absent, the step-by-step diagnosis walks up the chain of causes, and measuring your visibility says how to observe it without fooling yourself. To choose who helps you, our French guide choisir une agence GEO gives the criteria. The GEO Audit covers the six dimensions on your priority pages and implements the fixes.
Frequently asked
Do Google reviews influence AI visibility? No vendor documents it. We notice that answers about local businesses often pick up reviews and directory listings, and we do not make a rule of it.
Does Google Business Profile influence ChatGPT? Nothing documented either. A profile consistent with your site is good practice in itself; its effect on an assistant cannot be measured reliably today.
Is there a form to get listed on ChatGPT? No, and no assistant vendor offers an equivalent of Search Console. Access is set in your robots.txt and your CDN, the rest in your pages.