What is AI share of voice — and when is the number honest?
AI share of voice is the fraction of all brand mentions in AI answers about your category that belong to you. It is only meaningful when the field it divides by is defined and published — a share of an incomplete list overstates everyone on the list.
Updated 22 July 2026
Share of voice came to AI search from advertising, and it kept its shape: your mentions over everyone's mentions. Ask assistants the questions buyers ask, extract every brand named in the answers, and divide. Simple — and easy to get quietly wrong.
The denominator is the whole game
A share is a fraction, and the number that matters is the one on the bottom. Divide your mentions by a list of five competitors and you look dominant; divide by the forty businesses assistants actually name and the same mentions shrink to their honest size.
We know because it happened to us. When we first measured UK cigar retail, our defined field captured under half the merchant mentions the assistants actually made — meaning every share figure we could have shown was inflated by roughly two-to-one. The fix was not arithmetic; it was editorial: read every business the assistants named, decide by a stated rule which ones belong to the field, and publish the rule. Our methodology now states that criterion, and our first category report divides by a 39-merchant field anyone can check.
What a defined field changes
Two things, both worth money.
First, the numbers get smaller and truer. A brand that looked like 8% of the conversation may be 4% of the real one. If you are buying decisions with these figures — where to spend content and PR effort — the flattering version leads you to underinvest exactly when you should not.
Second, disagreement becomes possible. When someone asks "share of what?", an index with a published criterion has an answer: specialist merchants a buyer could choose, not marques, distributors or department stores. Without the rule, a share of voice is a share of whoever the tool happened to track.
Questions to ask any share-of-voice figure
- What is the denominator? A named, published field — or "the competitors you entered"?
- How were mentions counted? Name-matching needs an attribution filter; a brand with a generic name collects false positives without one.
- Across which engines? A share measured on one assistant says nothing about the others — they disagree far more than most people expect.
- With what confidence? Sampled answers are noisy; a serious measurement carries confidence intervals and refuses to read movement inside them as change.
The honest limit
A field is never finished. Assistants name new businesses as the web changes, and a real index keeps adding the ones that qualify — which means your share can fall without you losing a single mention. That is not the measurement failing; that is the denominator getting truer.
See where your category stands on the index, or run a free scan to start your own reading.