1. It writes its own searches, not yours
The first thing to understand is that ChatGPT almost never searches the question you asked. It rewrites your question into several of its own search queries, a process usually called query fan-out, and runs those instead. Ask it "best small group tours for retirees" and it will fire off searches like "best small group tours for retirees 2026 official senior travel companies" and "small group tours seniors accessibility reviews", often several at once.
Two things follow. First, it frequently seeds those searches with the brands it already has in mind, typing competitor names into the query before it has looked at anything. That tells you who the model currently treats as the field. Second, you are not competing for the words your customer types. You are competing to be findable for the rewritten queries the model generates, which are longer, more specific, and loaded with terms like "official", "reviews" and the current year.
2. It runs those searches on Bing, plus its own crawler
ChatGPT Search retrieves through Bing's index, plus OpenAI's own crawler, OAI-SearchBot. Both are required. If a page is not in Bing's index, ChatGPT cannot cite it. And if a page blocks OAI-SearchBot in its robots.txt, ChatGPT will not show it even when it ranks well on Bing (Yoast). OAI-SearchBot is separate from GPTBot, which OpenAI uses for training, so the two are handled differently.
The practical consequence is a visibility problem most brands never look at. You can be perfectly optimised for Google and still be invisible in ChatGPT because Bing has not indexed your key pages, or because your robots.txt quietly blocks the one crawler that matters. Claiming Bing Webmaster Tools, checking your pages are actually indexed, enabling IndexNow, and allowing OAI-SearchBot are the unglamorous prerequisites for any ChatGPT visibility at all.
3. It checks specific facts on the real sites
Once it has a shortlist, ChatGPT often runs narrow searches straight against the candidate websites to confirm details before it commits. We have watched it run, word for word, searches like site:example.com average group size and site:example.com small group travel 8 to 16 travelers. It is verifying the specific facts a buyer cares about, prices, group sizes, credentials, on the source site itself.
This is the most actionable part of the whole mechanism. If a competitor states its price or specification plainly and you bury yours in a PDF or an image, the model can confirm the competitor and not you. The facts your buyers ask about have to sit on your pages as clean, explicit, machine-readable statements, ideally marked up with structured data, or the model simply cannot use them.
4. It writes one answer and names a few sources
From everything it retrieved and verified, ChatGPT composes a single answer and cites a handful of sources. Being on that shortlist is the entire objective, because most people never click past the answer. For category questions, the sources it cites are usually not your own website. They are the third-party places it trusts for that topic: reviews, directories, forums, and publishers. Analyses of AI citations consistently find that the large majority come from third-party sources rather than brand sites.
5. How this differs from classic Google SEO
Traditional Google shows ten links and the customer clicks one, so the game is ranking higher. This is different. The machine reads the pages, verifies the facts, and writes one answer with no click. That changes what you optimise for. Instead of ranking a page for a keyword, you are trying to be three things at once: retrievable for the queries the model writes, verifiable when it checks your facts, and a clear enough entity that it puts your name into its own searches in the first place.
6. How to get cited by ChatGPT
The mechanism points to a short, concrete list. None of it is a trick; it is making your brand legible to a machine that reads the web on your buyer's behalf.
- Fix the plumbing. Claim Bing Webmaster Tools, confirm your pages are indexed in Bing, enable IndexNow, and allow OAI-SearchBot in robots.txt.
- Be present on the sources it pulls from. For most categories the answer is built from third-party sources, reviews, directories, forums, publishers, not your own site. Being cited there is most of the work.
- State your facts plainly and mark them up. Prices, specifications, credentials and answers to real buyer questions, in clean text and structured data, so the model can verify them.
- Fix your entity. A complete, connected Wikidata record, Organization schema with sameAs links, and consistent naming across the web help the model recognise you as a candidate worth searching for.
- Answer the real questions. Publish content shaped around the specific questions buyers ask, with FAQ schema, rather than brand-led marketing copy.
See what ChatGPT does for your brand
The quickest way to understand where you stand is to look. Open ChatGPT, Claude, Gemini, Perplexity and Google AI Mode and ask each a category question a buyer would ask. Count how often your brand appears versus your competitors, and note which sources get cited instead of you.
For a structured read across all the major AI platforms with a scored result, use our free AI visibility checker. You will see a score out of 100, a platform-by-platform breakdown, and the competitors AI recommends in your place. For the wider picture of why brands go missing, see our deep dive on why brands don't show up in AI generated answers.