What counts as a citation in AI search?

A citation and a mention are not the same event#

A citation is a source; a mention is a name, and Prefer counts them separately. When an AI engine answers a question, it can do two different things with your brand. It can name you in the text (“tools like Prefer and Profound do this”), which is a mention. Or it can list one of your pages as a source the answer was built from, usually as a linked reference or an entry in a sources panel, which is a citation. They look similar in a screenshot and behave completely differently in your reporting.

The two are measured separately for a reason. In our baseline ChatGPT run of 47 grounded answers, mentions and citations were tracked as independent numbers rather than folded into one score. The vendors who take measurement seriously do the same: Ahrefs Brand Radar’s own documentation defines the category by tracking mentions, citations and AI share of voice as distinct signals, not a single blended figure.

A citation is also engine-specific. It might surface as an inline link, as an entry in a “sources” list, or not at all. In the same baseline run, 3 short definitional prompts were answered straight from the model’s memory with no sources, so brands were mentioned with zero citations behind them. A citation is not always a clickable link; it is the engine pointing at where the claim came from, when it points anywhere.

Why the distinction matters#

Only a citation tells you which page earned the answer, so only a citation is directly actionable. If you optimize on mentions alone, you learn that you were named but not why, and you cannot tell what to change.

• Citations name the winning source. They point at the exact page or domain the engine trusted, which is the thing you strengthen, replicate or earn a place on.
• Mentions measure familiarity. They tell you the engine knows your brand exists in the category. Useful, but not a lever on its own.
• The gap is the strategy. When your mention rate is high and your citation rate is low, the engine knows you but trusts someone else’s page for the facts. Closing that gap, by making your own pages liftable or earning citations on the sources AI already reads, is where the work sits.

So the practical rule is to track both and watch the distance between them. The citation rate glossary entry covers how the metric is computed, and our study of 1,237 AI citations shows where citations actually land by source type. To see your own mention and citation rates across every engine, run a free AI visibility audit.

Sources

1. Prefer's run store tracks mentions and citations as separate metrics. On the baseline ChatGPT run (47 grounded answers), a mention (the brand named in the text) and a citation (the brand's domain present in the answer's sources) were counted and reported independently. (Prefer research note: AEO loop gap report (2026-09-02), dated research note)
2. Ahrefs Brand Radar's own product documentation defines the category by tracking mentions, citations and AI share of voice as distinct signals, and leads with a one-sentence definition of AI visibility across a 356M+ search-backed prompt database. Ahrefs: What is Brand Radar (help doc)
3. Not every answer carries a citation. In the same baseline run, 3 short definitional prompts were answered from the model's memory with no grounded sources at all, so brands could be mentioned with zero citations behind them. (Prefer research note: AEO loop gap report (2026-09-02), dated research note)
4. Prefer, our product, tracks mentions and citations of your brand as separate measures across ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode, on every plan. Prefer pricing (our product)