Mention rate, citation rate, and share of voice: the three numbers that matter

Three numbers cover most of what’s worth measuring in AI visibility. Mention rate: the share of tracked questions where AI names your business. Citation rate: the share where it credits a source about you. Share of voice: your slice of all the business names given across those questions. Everything else vendors report is usually one of these dressed up, or a number nobody can check.

The three metrics, defined to be checked

Mention rate is the percentage of tracked prompts where an AI assistant named your business in its answer. Citation rate is the percentage where the answer cited a source about your business, named or not. Share of voice is the percentage of all business names in those answers that were yours. All three come from counting real answers, which means all three can be verified by reading them.

That last sentence is the test that separates these from a “visibility score”: each of these numbers traces to recorded answers you could re-read. A composite score traces to a formula someone made up, which is why our methodology doesn’t produce one.

What each is good for, and where each is blind

Mention rate is the headline number: it answers “does AI recommend me.” Blind spots: it ignores how you were described (a lukewarm mention counts the same as a glowing one) and where in the answer you appeared.

Citation rate is the infrastructure number: it shows whether the sources about you are in the answer’s supply chain even when your name isn’t surfacing yet. A rising citation rate with a flat mention rate usually means the assistant is reading pages about you and not yet convinced, which is different, and more fixable, than not being read at all. The mention-versus-citation distinction is worth understanding before you watch either number.

Share of voice is the competitive number: it puts your mentions against everyone else’s. It’s also the easiest to game, because it depends entirely on which questions are in the set. Broaden the prompt set toward questions nobody wins and everyone’s share drops; narrow it toward your strengths and yours rises. When a vendor reports share of voice without publishing the prompt list, the number means whatever they need it to mean.

One business, ten questions, three numbers

An illustration, computed visibly so the math is the content. Say ten tracked prompts for a roofer, and across the ten answers: the roofer is named in 4, pages about the roofer (their site, a directory profile, a review page) are cited in 6, and the answers name 25 businesses in total, 5 of which are the roofer (one answer named them twice, which counts once per answer for mention rate but each name for voice).

Mention rate: 4 of 10, or 40%. Citation rate: 6 of 10, or 60%. Share of voice: 5 of 25, or 20%. Three different numbers from the same ten answers, and each says something the others don’t: recommended sometimes, read more often than recommended, and one name among many when the answers get written.

Which number to watch monthly

Our opinion: mention rate, and almost nothing else, until it’s stable. Share of voice matters once you’re mentioned regularly and the question becomes who else is; citation rate matters most when mention rate is stuck and you need to know whether the problem is being read or being believed. Watching all three daily is how owners talk themselves into panicking over normal week-to-week movement. One number, monthly, over many askings: that’s the discipline that survives contact with a busy schedule, and it’s the number your dashboard puts first for exactly that reason.