Both named in all four. Eleven points apart.
Counting is the first thing everybody does and the last thing that should be reported alone, because position inside an answer carries most of the value.
In a run we published, two brands were each named in four out of four answers. Identical counts. Their weighted share differed by eleven points, because one of them landed near the top of every list and the other did not.
That gap is not a rounding artefact. It is most of the value of being mentioned, and a report that counts appearances throws it away.
What counting misses
- Position. First named is chosen far more often than fifth named, and the difference is not small.
- Question value. A mention on a definitional query is worth less than one on a buying query, and a brand can look healthy entirely on the cheap ones.
- Denominator honesty. A count is only meaningful against the number of answers that could have named you, which failed calls and non-discriminating questions both distort.
- Name resolution. A brand split across variants shows several small counts instead of one real one.
What to report instead
Both, in the right order. The count first, because it is the figure anybody can verify by reading the answers, and the weighted share second, because it is the one that reflects what actually happens. A report that leads with the percentage is showing a number the reader cannot check.
The failure this prevents
A brand can hold its count steady while sliding down every list it appears in — same mentions, less business. Only the weighted figure catches that, and only the stored answers prove it happened.
The short answers
Why is counting AI mentions not enough?
Because position inside the answer carries most of the value. In one run, two brands were each named in four of four answers and their weighted share still differed by eleven points, because one landed higher in every list.
What should a share of voice figure be weighted by?
Position in the answer and the commercial intent of the question. Without the second, a brand can look healthy entirely on questions nobody buys from.
Should the count or the percentage be reported first?
The count, because a reader can verify it by reading the answers. The percentage second, because it reflects what actually happens but cannot be checked on its own.
How can my mentions stay flat while my visibility falls?
By sliding down the lists you appear in. Same count, lower position, less business — and only a weighted figure with the stored answers underneath will show it.
What distorts the denominator?
Failed calls counted as absences, and questions that named nobody at all. The first should be excluded from both sides with the coverage declared; the second means the question did not discriminate.
Find out in under ten minutes
One free check, no account and no card. You get the answers the assistants gave, not a score.
Every figure quoted here comes from a real question put to a real assistant, saved with the date and the assistant that answered it. This piece was written to answer the search "share of voice calculation".