Cited but Not Mentioned: Why AI Engines Use Your Page and Name a Competitor

If AI answers link to your website but never say your name, you have a ghost citation, and it is the most common outcome in AI search. Semrush's study of 126 million prompts found that 62% of AI citations never produce a brand mention. The model reads your page, takes the fact it came for, writes a confident answer, and credits a competitor it can describe in a sentence.

This is a strange kind of failure, because it looks like success from every angle except the one that matters. Your content is good enough to be trusted as a source. Your pages are indexed, parsed and used. And the buyer reading the answer has no idea you exist.

Why it happens

Three causes, in order of how often they turn out to be the real one.

Your page states facts and never states an identity

Models write answers out of quotable sentences. A page explaining how to choose a supplier is full of quotable sentences about choosing suppliers, and if not one of them says who you are, the model has nothing to attribute. It quotes the useful part and names a brand from somewhere else in its context, usually one whose own page said plainly "we are X, and we do Y for Z".

This is not about adding a call to action. A call to action is an instruction, and models do not quote instructions. It is about adding a fact.

Your identity is described differently in five places

Your homepage calls you a growth platform. Your LinkedIn calls you a marketing agency. A directory listing from 2023 calls you a social media tool. Each description is plausible and none accumulates enough weight for a model to state it with confidence. When a model is unsure what you are, it leaves you out of the sentence and keeps you in the footnotes.

Your content is informational when the prompt is commercial

Commercial prompts ask which, not what. Which tool, which alternative, which is better for a small team. Those answers are assembled from comparison-shaped sources, and an informational article, however good, is not one. Comparative content produces roughly 2.4 times more brand mentions than informational content because it already contains the sentence the model needs, with the brand name inside it.

Confirming it is really happening to you

Run the twenty-minute audit and record two columns rather than one: was the brand named, and was the site cited. Then compute the ratio of mentions to cited pages.

The signature is unmistakable once you look for it. A double-digit citation count with a low single-digit mention count means the problem is not reach and not quality. It is attribution. If instead both numbers are near zero, you have a different problem entirely, and the fix is off-site presence rather than on-site writing. The distinction is covered in how the score is built, because these two failures sit in different dimensions of it.

The fix that does most of the work

One identity sentence per cited page. Plain and factual: what you are, who you serve, what you do. Placed early, written as a statement rather than a pitch. It is a small edit and it resolves more ghost citations than everything else combined, because it addresses the cause directly: the model had a fact and nothing to attach it to.

Write it as something a stranger could quote without knowing anything else about you. "Northgate Dental is a family dental practice in Leeds offering same-day emergency appointments" works. "We put your smile first" does not, because there is nothing in it to attribute.

Start with the pages that are already being cited rather than with the whole site. Those are pages a model has already decided to trust, so turning ten trusted pages into ten attributable pages is a week of work with a compounding return, and it beats publishing twenty more that will be cited anonymously too.

Everything else that helps here is the same work that gets you named in the first place: answer-first structure, comparative content, one consistent description off-site, reviews, structured data and crawlability. Those are covered in order of return in the seven signals that decide who gets named, and there is no reason to state them twice.

What to expect, and when

Give it 30 to 60 days. Pages need recrawling and answers refresh on their own schedule, so nothing you do this week shows up next week. Watch the mentions-to-citations ratio rather than the headline score, because the ratio responds first: the same pages start producing the same citations with your name attached.

Do not rewrite the whole site. Start with the pages that are already being cited, since those are pages models have already decided to trust. Turning ten trusted pages into ten attributable pages is a week of work with a compounding return, and it is strictly better than publishing another twenty articles that will be cited anonymously too.

Tracking both numbers

Markty AI scans ChatGPT, Gemini and Perplexity four times a month and reports mentions and citations as separate figures, alongside the queries you appear for and the ones where competitors are named instead. The gap between those two numbers is the most honest measure of whether your content is working for you or for somebody else.

Being cited is being used. Being mentioned is being recommended. The whole discipline of AI visibility is the work of converting the first into the second.