Assistants do not read a page the way a customer does. They lift the part that answers the question cleanly, and the research says they usually find it near the top or not at all.
How to get cited by ChatGPT
To be cited by ChatGPT, put the direct answer in the first third of the page, write each claim so it can be lifted without the paragraph around it, keep your name and details consistent everywhere the model has already seen you, and let OpenAI's search crawler read the page. Those four things cover almost everything a business actually controls. What you do not control is larger: the assistant decides what the question means, breaks it into sub-questions you never see, and assembles an answer from whichever pages resolve those sub-questions cleanly. So the goal is not to rank for a phrase. It is to be the easiest page in the set to quote.
Nearly half of citations come from the first third of the text
An analysis by the growth advisor Kevin Indig, reported by Search Engine Land, examined 18,012 verified ChatGPT citations pulled from a much larger body of AI answers and found that 44.2 percent of them came from the first 30 percent of a page's text, with the likelihood of being quoted falling steadily from there. Be careful about what that does and does not mean. It does not mean the rest of the page is wasted, since the depth underneath is usually what makes the opening credible enough to trust in the first place. It means extraction happens near the top, and almost all business writing is built the other way around. The standard agency article warms up for four paragraphs, establishes the problem, builds a little tension, and delivers the genuinely useful sentence somewhere near the close. That shape was designed for a reader who scrolls and rewards patience. It is close to the worst possible shape for a system that lifts.
Ranking well and being cited are different games
Ahrefs analysed roughly 15,000 queries and found that only about 12 percent of the URLs cited by AI assistants also appeared in Google's top ten organic results for the same query. That figure is easy to misread. It does not mean rankings stopped mattering. It means the two systems are resolving different questions. Google ranks pages against a query. An assistant assembles a response out of fragments that answer the pieces it has broken that query into, which is why a page sitting fourth can be quoted while the page sitting first is passed over, its one useful claim buried inside a case study three screens down. For a smaller company the practical consequence is unexpectedly encouraging. You do not have to outrank the incumbents to be quoted next to them. You have to be more quotable than they are, and most of them are not trying.
How to get cited by ChatGPT: the four things you control
Start with structure. Every page should answer its own title within the first hundred words, in plain declarative sentences, before any context or qualification arrives. Giving away the conclusion first feels wrong to anyone trained to build an argument, but the reader who came for it is better served, and the model looking for a quotable line has found one.
Then make the claims extractable. A sentence that depends on the two paragraphs above it to make sense cannot travel, and anything that cannot travel cannot be cited. That means naming the subject rather than saying it or this, attaching the number to what it measures inside the same sentence, and citing sources by name in the text rather than as a footnote, since attribution the model can see is attribution it can pass on.
Third, be a consistent entity. Assistants build an impression of a company from every place it appears, so a business described one way on its own site, differently on its profiles, and under a slightly different legal name elsewhere is harder to identify confidently and therefore riskier to quote. One description, one name, one set of details, everywhere.
Fourth, check access, because none of the above matters if the page cannot be fetched. OpenAI publishes three separate crawlers with three separate purposes: GPTBot collects training data, OAI-SearchBot surfaces and links to pages inside ChatGPT's search results, and ChatGPT-User fetches a page on demand when a user or a custom GPT asks for it. They are controlled independently in robots.txt, which means a site can decline to feed model training while still allowing itself to be cited. Blocking all three by reflex, which plenty of sites did, removes you from the answers entirely.
See how we structure content so it ranks in search and gets quoted in AI answers.
A working checklist for the next page you publish
Six checks, in the order that saves the most rework:
- Answer the title in the opening paragraph in fewer than a hundred words, then delete the introduction you wrote before it, because the warm-up is almost always the part that pushed the answer out of the zone where it gets read.
- Read each key sentence aloud on its own and ask whether a stranger would understand it without the paragraph around it, since that is the exact test a model applies when deciding what it can safely lift.
- Attribute every third-party number in the text itself, naming the organisation and what was measured, because an unattributed statistic is both less citable and less honest, and the two failures have the same fix.
- Give each substantive question its own heading phrased the way a person would ask it, so the page presents several distinct answerable units rather than one long undifferentiated argument.
- Confirm in robots.txt which crawlers you allow, deciding deliberately between training access and search access rather than inheriting whatever your platform set by default.
- Track impressions and citations separately from clicks, because a page that is quoted without being visited is still doing work, and judging it on sessions alone will lead you to delete the content that is building your reputation.
Getting cited by AI assistants: common questions
- How do I get my website cited by ChatGPT? — Answer the page's own question in the first third of the text, write claims that stand alone without surrounding context, keep your company details identical across every profile and directory, and allow OAI-SearchBot in robots.txt so the page can be fetched for search results.
- Does ranking on Google mean ChatGPT will cite me? — Not reliably. Ahrefs found only about 12 percent of AI-cited URLs also appeared in Google's top ten for the same query, because assistants quote fragments that answer sub-questions rather than ranking whole pages against one phrase.
- Should I block GPTBot? — That depends on whether your objection is to training or to visibility. GPTBot collects training data while OAI-SearchBot handles search and citation, so blocking the first and allowing the second keeps you quotable without contributing content to model training.
- Where should the answer go on the page? — As close to the top as the sentence can be made to make sense, since analysis of 18,012 ChatGPT citations reported by Search Engine Land found 44.2 percent came from the first 30 percent of the text.
Keep reading
- Answer-engine optimization: how to become the answer, not just a link
- SEO services in Malaysia: what you are actually paying for
- Measure what compounds: analytics for decisions, not dashboards
Talk to us about building content that earns citations instead of renting clicks.
Build something
worth trusting.
Tell us what's slowing your business down. We'll show you the system that fixes it — and how fast.