Short answer

The mention rate is the percentage of valid answers, to unbranded questions, that name your company. Share of voice in AI is how those mentions split between your brand and the agreed competitors, within the same sample. Both describe a sample of chosen questions, on certain platforms, on certain dates. Neither is market share or the probability that a buyer will see you. Without the numerator, the denominator and the matrix behind them, they are empty figures.

Mention rate

Numerator: valid answers that mention the brand. Denominator: valid answers to unbranded questions. Errors (timeouts, blocks, empty answers) are excluded from the denominator by a rule set before looking at the results, and are counted separately.

Example: 12 questions on 2 platforms make 24 observations. 2 returned an error. Of the 22 valid ones, 8 mention the brand. Mention rate: 8 of 22, or 36%. You publish “8 of 22”, not just “36%”.

A mention is not a recommendation. “We do not recommend X for freelancers” mentions X. That is why the recommendation rate is measured separately: answers that explicitly recommend the brand.

Share of voice in AI

You count the mentions of each agreed brand in the same sample and split them. If, across the 22 valid answers, your brand appears 8 times, competitor A 14 times and competitor B 6 times, the total is 28 mentions and your share of voice is 8 of 28, or 29%.

Two warnings. First: you choose the set of competitors before measuring; changing it afterwards changes the figure. Second: one answer can mention several brands, so share of voice does not say who “wins” an answer, but how often each brand appears in total.

Position

Only when the answer contains a clear list. If it is a narrative paragraph, no ranking is forced onto it. A report’s “average position” should say how many answers with a list it was calculated on.

Why it is not market share

The sample is twelve questions that someone chose, on two platforms, on one day. It is not a representative survey of what all buyers ask, or of what they see. Saying “we have 36% AI visibility” and nothing more turns a sample into a claim about the market. The honest version is “in this sample, on these dates, it appeared in 8 of 22 answers”.

How to read them properly

  • Always look at the denominator and the errors.
  • Look at the matrix: a stable rate can hide the fact that almost every row changed.
  • Only compare with the same sample: same questions, same platforms, another date.
  • Be wary of a single figure presented as a score.