A citation is a link to your site as the source behind an AI answer. A mention is your brand name appearing in the text of that answer. CitedScore tracks them as separate numbers in every audit: citation rate, mention rate, and ghost rate, the share of your citations that never say your name. They get three numbers because they rarely move together.
How does CitedScore tell the two apart?
Every answer in an audit lands in exactly one of four buckets: Both, Mentioned, Cited only, or Absent. An audit asks 20 buyer questions, built from four buyer personas plus a set of neutral and head-to-head questions, on four platforms (ChatGPT, Google AI Overviews, Gemini, and Perplexity), for 80 observations. Nothing gets rounded into a single "visible" catch-all, because a citation with no mention doesn't behave anything like a mention with no citation from a buyer's point of view.
Picture a landscaping company whose page gets linked in an answer about clay soil drainage, with no mention of the company's name anywhere in the text. That's Cited only. The same company's name might show up in a separate answer about trusted local landscapers, with no link back to their site. That's Mentioned. A third prompt might return neither a link nor a name at all, Absent, usually meaning the model didn't consider that page relevant enough to draw from for that specific question. Four buckets, and most sites land unevenly across them rather than clustering in one.
Why do citations and mentions diverge instead of moving together?
They come from two different decisions inside how an AI engine builds an answer. A citation gets decided when the engine chooses which page backs up a specific claim. A mention gets decided separately, when the engine decides your brand name belongs in the text it's writing. A page can win the first decision and lose the second entirely, and that split shows up constantly across real audits. That gap has a name: a ghost citation.
Where does this classification get shaky?
A single prompt run once isn't reliable enough to act on. The same question asked twice can return a different citation or mention result, since AI answers aren't fully deterministic and models get updated on their own schedule. One favorable answer doesn't establish a pattern, and one unfavorable one doesn't either. This is the main reason a screenshot of one good answer, the kind that circulates internally as proof a strategy worked, doesn't hold up as evidence on its own.
That's why CitedScore runs the classification across 80 observations per audit instead of one. Aggregating catches the pattern instead of a single lucky or unlucky draw. Share of voice only gets reported for a platform once your brand and your competitors have been mentioned often enough to make the comparison meaningful. Below that threshold, CitedScore holds back rather than publishing a small-sample number that would mislead more than it helps.
How often should you recheck your split?
About once a quarter for most sites, since models update on rolling schedules nobody outside the AI labs controls, and a split that looked strong three months ago can shift without anything on your site changing. A site in a fast-moving category, one where competitors are actively working on their own AI visibility, benefits from checking more often than that. A single audit is a snapshot of one moment, and ongoing monitoring is how you see the trend: whether a fix landed, and whether a model update moved you.
Which one should carry more weight when you're evaluating a site?
Both count, and CitedScore combines them. The GEO Citation Score, CitedScore's composite metric, rolls both signals into one number from 0 to 100, the way a credit score rolls several signals into one. Mentions carry more of the weight than citations, and a brand mentioned early in an answer counts for more than one mentioned near the end. Each platform's result is then weighted by how many buyers it reaches, so ChatGPT and Google AI Overviews count for more than Gemini or Perplexity.
That weighting is why a site that gets cited often but mentioned rarely can score below a site with fewer citations and more mentions, even though the first site does more of the analytical work. GEO visibility takes both: the citation proves your content earned its place in the answer, and the mention makes sure the buyer knows whose content it was. The composite rewards balance, because a citation nobody sees credited does less for a buyer standing in front of a purchase decision than a mention that puts the name directly in the text they're reading.
A site with a strong citation rate and a weak mention rate is doing real analytical work for AI answers while getting almost none of the buyer-facing credit for it. See the full breakdown of what that costs, which lever to prioritize fixing first, or how to check your own split by hand.
See one answer from each platform before you guess at your split. Run a free scan: one real buyer question, four platforms, about a minute.
Run a free scan