How to get your brand mentioned by ChatGPT

SEO is only part of it

“Surely this is just SEO” is a reasonable objection. Search engines already reward useful pages, reputable sites and clear answers, so improving those things should help here too. They do help. A page that is vague, difficult to read or hidden from crawlers has little chance of appearing anywhere useful. But SEO answers a different question from the one AI engines are asking, and that gap is where many brands lose the mention.

Start with the event you actually want.

For the question of how to get your brand mentioned by ChatGPT, the important split is between a mention and a citation. A mention is your brand being named in the answer. A citation is a source link used to support it. They overlap, but they have different causes. Mentions often come from other sites discussing your product, while citations often come from a page that answers the buyer's question directly.

That difference changes what you measure. A brand can be named without its site being cited, and its site can be cited without the answer naming the brand.

Mentions usually come from elsewhere

For brand mentions in ChatGPT, begin with the pages that already discuss products in your category. In the recorded research sample, listicles and roundups account for 67.5% of citations, while competitor product pages account for 24.4%. Together, they account for 91.9% of citations. Those pages give an AI answer the context it needs to place products alongside alternatives.

Your own page is not where this argument gets made.

That does not mean you should pursue every listicle or ask for a favourable sentence. It means your site is rarely the only place that establishes what your brand is, who it serves and which alternatives belong beside it. Relevant comparisons, independent reviews, community discussions and category pages can all create the context from which a mention becomes likely.

Look for missing context before chasing volume. If third-party pages describe your product inconsistently, fix the underlying facts first. If they never place your product in the buyer's comparison set, publishing another product page on your own domain may not solve the problem.

Citations usually answer questions

Citations have a different job. When an AI answer links to your page, that page needs to answer the question the user asked, not merely describe your company. A useful page makes the relevant claim easy to find, supports it with specific information and keeps its scope clear enough for an engine to use.

The page needs a job.

The wider CiteGraph research recorded 45,309 citations across 7,260 AI answers, from 60 scans of 41 sites in 44 categories. Those answers asked 442 distinct buyer questions and cited 16,506 distinct pages across 7,624 domains. The scale is large enough to show patterns, but it does not turn every individual answer into a rule.

The page types matter more than a generic instruction to create content. Listicles and roundups led the sample at 67.5%, competitor product pages followed at 24.4%, while the scanned product's own site accounted for 1.5%. Your own site still matters, but it has to give the engine a clean, quotable answer to the question behind the search.

A page built for citation should state what the product does, who it is for, where it differs, what it costs if that is relevant to the question, and what evidence supports those claims. Plain prose is easier for an engine to lift; prose that requires assembly usually isn't.

Mentions and citations move independently

The engine sample shows why a single visibility number is not enough. Across 1,781 sampled ChatGPT-family answers, the scanned product was named in 4.4% of them, with 3.4 sources cited per answer on average. Perplexity named the scanned product in 8.2% of 1,787 sampled answers and cited 5.3 sources per answer. Gemini named it in 6.8% of 1,784 answers and cited 4 sources. Claude named it in 6% of 1,467 answers and cited 4.8 sources.

One score can hide the cause.

These figures are not a ranking of product quality. They describe different samples and different answer behaviour. They do show why a useful report needs separate fields for whether the brand was named and whether its pages were cited. A rise in citations can happen without a rise in mentions. The reverse can happen too.

For a practical report, record the buyer question, the engine, whether the brand was named, and which source pages were cited. Third-party coverage belongs in the mention workstream. On-site answer quality belongs in the citation workstream. A blended score hides which lever moved.

CiteGraph's own visibility checker and methodology page show how this split is measured in practice. The method matters because an answer count without its question set and sample boundaries tells you very little.

The sample has real edges

Here is a worked example using the recorded figures. Line one: 1,781 ChatGPT-family answers were sampled. Line two: the scanned product was named in 4.4% of them. Line three: those answers cited 3.4 sources per answer on average. The figures describe naming and source volume, not the same event.

The denominator matters.

Line four: across the wider sample, the scanned product's own site accounted for 1.5% of citations. That is a share of source links, not a share of ChatGPT-family answers. You cannot subtract 1.5% from 4.4%, or treat the result as the number of answers that named the product without citing its site. The available figures do not match those denominators.

Line five gives the practical reading. ChatGPT named the product in 4.4% of sampled answers, while the wider citation data shows that own-site pages made up 1.5% of recorded citations. Those figures point to separate jobs: build relevant external context for mentions, and make your own pages useful sources for citations.

Line six keeps the wider sample in view. Listicles and roundups supplied 30,585 citations, competitor product pages supplied 11,052, and the product's own site supplied 663. The distribution is not subtle. It is also not a guarantee for any particular category or question.

A sample cannot promise inclusion

You cannot guarantee a ChatGPT mention by changing a title tag, adding a schema field or publishing a batch of articles. Those actions may improve the conditions for being found or understood. They do not control which sources an engine selects for a particular answer.

The result has boundaries.

Our data covers a few dozen scans, not the whole market. CiteGraph's own weekly research describes a measured sample, not every category, site or conversation. If third-party pages do not name you, work on category presence and accurate comparisons. If they name you but your pages are not cited, work on direct answers and supporting evidence.

We measure samples; we print margins; we do not know what we have not measured.