ChatGPT cites a handful of pages per answer. Here is how it finds them, what it quotes, and the six changes that make your site one of them.
How ChatGPT decides which pages to cite
When ChatGPT answers with citations, it is not recalling your page from training. It is running a web search behind the scenes, reading a short list of results, and writing an answer from the passages it found useful. The citations are the pages whose passages made it into the answer. That means getting cited is a retrieval problem first and a writing problem second.
Three things have to go right in sequence. Your page has to be in the index the model searches and reachable by its crawler. It has to be returned for the question the user typed, which is usually longer and more specific than a Google query. And it has to contain a passage that answers the question clearly enough for the model to use it without rewriting. Miss any one of the three and a competitor gets the citation.
The useful consequence is that none of this is mysterious. Each step is checkable, and most of the fixes are ordinary web work done with a slightly different reader in mind.
Step one: let the crawler in
OpenAI runs more than one crawler, and they do different jobs. GPTBot collects pages for training. OAI-SearchBot fetches pages for search results inside ChatGPT. ChatGPT-User fetches a page when a user asks about it directly. Many sites blocked GPTBot in 2023 to stay out of training data and never revisited the file, and some rules written then also catch the search crawler by accident.
Open your robots.txt and read it as those three bots would. If you want citations, OAI-SearchBot and ChatGPT-User need to be allowed. Then check that the page itself is readable without JavaScript: a crawler that gets an empty shell and a loading spinner has nothing to quote. Finally, look at bot protection on your CDN or firewall. Rules that challenge unknown user agents can block the search crawler without ever showing up in your analytics.
Brainito's AI SEO readiness check runs these access tests for a site and lists what is blocked, which is faster than reading headers by hand.
Step two: write passages the model can lift
A citation is almost always a short passage, typically one to three sentences, that states a fact, a definition, a number or a recommendation plainly. The model is looking for the sentence it can paraphrase with confidence. Pages that bury the answer under a story, a warm-up paragraph and three qualifications give it nothing to hold.
Rewrite the sections that matter so the first sentence under each heading is the answer. Define terms in a sentence that starts with the term. Put numbers in the text, with the unit and the source, rather than in an image. Use headings that are the question people actually ask, and keep each section about one thing so a passage does not need the paragraph above it to make sense.
- Answer first. "A marketing audit is a structured review of a company's marketing across channels" beats "In today's fast-paced world, businesses are asking themselves...".
- One claim per sentence. Compound sentences get split or dropped.
- Specifics over adjectives. "Plans from $20 a month" is quotable. "Affordable pricing" is not.
- Name the source. A number with a source is more likely to be used than a bare number.
Step three: make your entity unmistakable
Models hesitate to cite a source they cannot place. If your company name, what you do and who you serve are described three different ways across your site, your social profiles and the directories that list you, the model has a weaker picture of you than of a competitor with a boring, consistent description everywhere.
Fix the basics. Write an About page that says in one paragraph what the company is, where it operates and what it is known for. Add Organization schema with the same name, logo, address and the sameAs links to your profiles. Use one description of the business and reuse it on every third-party listing. Then earn mentions on pages the model already trusts: industry directories, review sites, trade publications and the forums where your customers talk. Third-party corroboration is what turns a claim on your site into a fact the model is willing to repeat.
Step four: cover the question, not the keyword
People type two or three words into Google and a full sentence into ChatGPT. "best crm small business" becomes "what is the best CRM for a small plumbing company with two office staff and no IT person". The second question has constraints a keyword page never addresses. Pages that get cited for it are the ones that talk about company size, team setup and the trade-offs between options, in plain prose.
The practical move is to collect the real questions. Ask your sales team what prospects ask on calls. Read your support tickets. Type your topic into ChatGPT and look at the follow-up questions it proposes. Then build pages or sections that answer those questions one at a time, with the comparison tables, the pros and cons and the "it depends" cases spelled out. The pages do not need to be long. They need to be complete for the question.
Step five: measure citations, then iterate
You cannot manage this from a rankings report. The measurement is simple to describe and tedious to do by hand: take a set of twenty to fifty prompts your buyers would ask, run them through ChatGPT on a schedule, and record which domains get cited and whether yours appears. Over a month you will see which questions you own, which a competitor owns, and where nobody consistent is cited yet. The gaps are where a well-written page wins fastest.
Brainito's AI SEO module automates the loop: it suggests prompts from your business and pages, asks ChatGPT, Perplexity, Gemini and Google's AI Overviews each week, records the citations and competitor share, and turns every blocker and content gap into a task. If you would rather start smaller, the free marketing audit includes AI visibility as one of its 77 factors and shows whether the basics above are in place.
Getting cited by ChatGPT is not a trick and it is not a separate discipline from good SEO. It is the same work, judged by a reader that quotes rather than ranks. Let the crawler in, write the sentence it can lift, make your name consistent, answer the whole question, and check the result every week.