Mentions vs citations: the two numbers that matter in AI answers
AI visibility collapses into two distinct events, and conflating them produces bad strategy.
A mention is your brand named in the answer’s prose — in text and entities[]. The user reads it.
A citation is your domain in sources[] — the engine used your page as evidence. The user can click it.
Why the split matters
You can be mentioned without being cited (the model knows you from training data or another source describes you) and cited without being mentioned (your page informed the answer but you weren’t named). Each combination needs a different response:
| State | Meaning | Fix |
|---|---|---|
| Mentioned + cited | Ideal — you’re the recommendation and the evidence | Defend: keep the cited page fresh |
| Mentioned, not cited | Known but not sourced — risky, competitors’ pages tell your story | Publish citable evidence pages so the citation follows the mention |
| Cited, not mentioned | You informed the answer but got no credit | Make the claim about you on the page — engines cite what states names |
| Neither | Invisible | Create the content the fan-out queries surface |
Measuring both
Every monitor response gives you the raw material: scan entities[] for mentions, sources[] for citations. Per prompt set:
mention_rate = prompts naming you / total
citation_rate = prompts citing your domain / total
gap = mention_rate − citation_rateA positive gap means reputation without evidence — fragile, since the engine learned about you through someone else’s lens. A negative gap means you’re the trusted source for recommendations going to competitors — which is worse.
The practice
Run your prompt set weekly, track both rates per engine and market, and watch the gap. Content strategy follows the quadrant you’re in — see the share-of-voice framework for the full metric set and citation selection for the mechanics underneath.