How to get cited by ChatGPT, and how to get your brand mentioned first

To get cited by ChatGPT, get mentioned across the web first, then make the page easy to quote. The model assembles a shortlist of sources it already associates with your topic, and only then decides which passage to lift and attribute. Most teams work on that second step and skip the first, which is why their pages are extractable and still never named.

The evidence for that order is unusually clear now, and it is not what the SEO reflex expects. The signals that best predict whether a brand shows up in AI answers are almost all off-site, and the strongest of them are mentions rather than links. Here is what the data says, and what to actually do about it.

How does ChatGPT decide what to cite?

Two stages, and they fail for different reasons. First, retrieval: the model, or the search layer behind it, gathers candidate sources for the question. Second, synthesis: it writes an answer and attributes specific claims to specific pages. Failing the first stage means you were never considered. Failing the second means you were considered and your page had nothing liftable in it.

Brand and topic association drives stage one. Page structure drives stage two. Both matter, but they are not interchangeable, and the fix for one does nothing for the other.

Brand mentions beat backlinks by roughly three to one

Ahrefs analyzed 75,000 brands to see which signals predict inclusion in Google's AI Overviews. The top three were all off-site brand signals, and classic backlinks came a distant fourth.

That study measured AI Overviews rather than ChatGPT specifically, so treat the exact numbers as directional. The pattern itself is consistent across engines and easy to explain: a language model learns that a brand belongs to a category by seeing the name written next to that category thousands of times. A link is a machine-readable vote. A mention is a sentence a human wrote about you, and sentences are what the model was trained on.

The practical consequence is that an unlinked mention in a comparison article you did not commission is worth more than a link from a directory nobody reads. It also means the outreach that works here looks less like link building and more like being genuinely present in your category: review sites, comparison roundups, industry publications, Reddit threads where someone asks what to use, and video, which the correlation above suggests punches well above its weight. Teams that get this right are usually not producing five times more content, they are just turning one post into every channel instead of letting it live and die on their own blog.

Most categories still have no owner

The other useful finding is how open the field is. Semrush tracked more than 50,000 brands across 1,094 US topics between January and June 2026, covering 600,000 citations and 220,000 URLs. Only 15.2% of those topics had a clear brand owner. Another 31.2% had an emerging leader. The remaining 53.7% were unsettled, with no brand dominating the answers at all.

Better still, the pattern runs against size. Only 11.3% of high-volume topics had a clear winner, compared with 19% of low-volume ones. And the classic authority metrics predicted topic ownership barely better than a coin flip: higher organic traffic described 48.4% of winners, higher authority score 52.5%. Being the biggest domain in your space does not reserve you a seat.

Two cautions come with that. Once a brand does establish itself it is sticky, holding first place in 90.4% of month-to-month comparisons, so the window closes. And there was only a 21% correlation between the most-cited domains and the most-mentioned brands, which is a reminder that being quoted as a source and being recommended as a product are two different wins. You want the second one.

What to fix on your own pages

Stage two is entirely in your control, and it is the cheaper half. A model cannot quote a sentence that only makes sense after three paragraphs of setup, so the work is making every important answer survive being cut out of its surroundings.

1. Answer in the first 40 to 60 words. Directly under the heading, self-contained, no setup required. This is the single highest-return change on most pages.
2. Use the buyer's question verbatim as a heading. Not a clever paraphrase. Matching a query to a passage starts with matching the wording.
3. Put comparisons in real tables. Rows and columns come through an extraction step intact. A beautifully written prose comparison does not.
4. Give every number a date and a source. A statistic with attribution is far more quotable than the same statistic floating alone, and it is what makes a model comfortable naming you.
5. Show an honest last-updated date. Freshness carries more weight here than in classic search, and an undated page reads as stale to both readers and models.
6. Let the AI crawlers in. GPTBot, OAI-SearchBot, ChatGPT-User, PerplexityBot, ClaudeBot and Google-Extended all need to be allowed in robots.txt. One careless disallow removes you from every engine at once.

None of this costs you anything in classic search, which is the reassuring part. Answer-first writing improves engagement, question headings match long-tail queries, and tables earn featured snippets. If you want the full version of the on-page layer, the answer engine optimization page covers every answer surface, and generative engine optimization goes deeper on AI chat specifically. Running these checks by hand across a whole site is exactly the kind of job worth handing to SEO automation tools.

You still have to rank first

Around 92% of pages cited in AI answers already rank in the top 10 for the related query. Generative engines do not crawl the open web from scratch when you ask them a question; they retrieve from a pool the search layer already trusts, then synthesize. If your page sits on the third results page, none of the formatting work above will reach it, because it is not in the pool.

So the order of operations is unglamorous and worth stating plainly. Rank the page. Then make it quotable. Then build the mentions that get your name into the shortlist for questions where no single page is the answer. Skipping straight to the third step is how teams end up with a brand-monitoring dashboard, a lot of activity, and no citations.

How to check whether any of it worked

There is no Search Console for ChatGPT, so measurement is sampling. Write down the ten questions a buyer asks before they know you exist, phrased the way they would actually type them, and avoid brand-name prompts, which flatter you and predict nothing. Ask each one in a temporary chat with memory turned off, because history will quietly feed your own brand back to you.

Log three things per prompt: were you mentioned, were you linked, and who was named instead. That competitor column is the useful one, because it tells you the specific page you have to out-answer. Repeat monthly with identical wording, and cross-check referral traffic from chatgpt.com, perplexity.ai and gemini.google.com in analytics to confirm citations turned into visits.

When a question keeps going to someone else, the answer is usually on the page rather than in the tool. The AI visibility tracker comparison covers what the dedicated monitoring tools do and what they cost, and what an AI visibility score actually measures explains why two dashboards give the same brand different numbers.

The short version

Citations follow mentions, and mentions follow being genuinely present in your category. The on-page work decides which of the shortlisted pages gets quoted, and it is quick, so do it. The brand work decides whether you make the shortlist at all, and it is slow, so start it now while more than half of US topics still have nobody sitting in the chair.

Questions people ask about getting cited by ChatGPT