Citations vs Mentions: The Two Kinds of AI Visibility and Why Both Matter

A mention is a name in the text

A mention happens when the answer says your name. "Popular options include Acme, Beta, and Gamma." No link, no source, just the name in the sentence. Mentions tend to come from what the model already knows, from training data or from the overall shape of what it read during a search. They are a signal that the model associates you with the category.

Mentions are what most buyers see. Many people never click a citation. They read the list of names, maybe ask a follow-up, and go search the one that sounded right. If you are not in the list, the citation on page three of the sources panel does not help.

A citation is a link to a page

A citation happens when the answer points to a specific URL as support for a claim. Perplexity numbers them. Copilot numbers them. Claude and ChatGPT show them inline or in a sources list when search is on. Google AI Overviews show them as cards.

Citations tend to come from the search step. The model fetched your page, used something from it, and attributed it. They are a signal that a specific page is readable, relevant, and quotable. They can appear even when your name is not in the answer text, for example when a page of yours is cited for a definition or a statistic and the answer never says who published it.

They come from different work

Mentions are mostly an entity problem. The model needs to know you exist, what category you belong to, and what distinguishes you. That knowledge comes from consistency across many sources over time: your site, your schema, your llms.txt, your review profiles, your presence in comparison lists and community threads. There is no single page that fixes mentions.

Citations are mostly a page problem. A specific page needs to be crawlable, state the answer plainly, and match the question closely. You can earn a citation for a page published last month if it is the clearest available answer. The work is on-page and it is fast.

Companies that are strong in one and weak in the other are common. A well-known brand with a bloated website gets mentioned but not cited. A sharp technical blog from an unknown company gets cited but not mentioned. Each case has a different fix.

Count both, separately

For a fixed prompt set, record for each answer: was the company named, and was any company page cited. Compute a mention rate and a citation rate. Track them per engine, because the engines differ. Perplexity tends to cite heavily and mention selectively. Models answering from memory tend to mention and not cite at all.

Then look at the combinations. Mentioned and cited is the goal. Mentioned but not cited means your pages are not the source the model is using to talk about you, and someone else's description is winning. Cited but not mentioned means your page is useful but you have not registered as an entity. Neither means you are not in the conversation.

Share of answer, the way AEO GEO Labs reports it, is built on these two rates plus a third: the share of cited sources you appear in. The third number tells you whether the pages the engines rely on, the lists and review sites and forum threads, include you.

Reading the gap

The most actionable finding in most audits is the mentioned-but-not-cited gap. It means the model trusts a third party to describe you more than it trusts you. The fix is a page written to be quoted on the exact question, with the facts stated in the first paragraph. The second most common finding is the reverse, where good pages exist but the entity is weak. The fix there is consistency work across every profile.

Run twenty prompts and build the two columns yourself. Or let AEO GEO Labs run the first audit and show you where each gap sits, with a weekly read so you can see which fixes moved which number.