Share of voice

Share of voice is the number most AI visibility tools lead with, and the one most often misread. In Attensira it is a rate: of the answers an engine gave to your questions, the fraction that named your brand. It is not your slice of a pie, and the difference changes how every chart of it should be read.

How is share of voice calculated?

share of voice = answers that named your brand
÷ all successful answers in the window

The unit is a run: one question, asked of one engine, in one country, at one time. An answer that names you nine times counts once, the same as an answer that names you once. A link is not needed; a citation to your domain is counted separately, as citation rate. Runs that fail are stored but left out of the denominator, so an engine outage shrinks the sample instead of lowering your rate.

Every competitor gets the same calculation over the same denominator. That single choice is what makes the numbers behave differently from the advertising version of the term.

What does share of voice look like on real data?

Here is our own share of voice. Attensira tracks 61 questions about its category, asked of ChatGPT and Perplexity every day. In the 30 days to 4 October 2026 the engines gave 4,524 answers, and this is how often each brand was named in them, read over the Attensira MCP that evening: Semrush in 1,621, Profound in 1,420, Ahrefs in 1,171, Peec AI in 988, Otterly AI in 967 and Attensira in 197.

Add the six rates up and you get 140.7%. Nothing is wrong. An answer to "best AI search visibility tools" routinely names five products, and that one answer lifts five rates at once. The numbers describe something true: Semrush is named in more than one answer in three, and we are named in about one in twenty-three. We publish our own low number because a share of voice you can only show when it flatters you is not a measurement.

The data covers only the two engines this workspace tracks, and a rate always comes with the sample it was read from; this one is n = 4,524.

Why does Attensira not make share of voice add up to 100%?

We could divide your mentions by every tracked brand's mentions. The total would then be 100%, and the chart would look like the advertising one. We do not, because the denominator would be the competitor list you typed into settings. Add a rival and your share falls; remove one and it rises. Nothing in the answers would have changed. A number that moves when you edit a settings page is measuring the settings page.

The consequence: a share of voice from a tool that does compute the pie version is not comparable to ours. Ours will usually look larger, because it answers a different question: "in how many answers were we named?" rather than "what fraction of brand mentions were ours?"

What do people get wrong when they read it?

• Reading the percentage before the sample size. Every rate arrives as {value, n}. A rate of 33% at n = 3 is one answer out of three; the next answer makes it 0% or 67%.
• Treating "not measured" as zero. An empty value means nobody asked that engine. A zero with n above zero means it was asked and did not name you. Charting the first as the second invents a rejection that never happened.
• Calling noise a change. Attensira tests every movement with a two-proportion z-test at 95%. A movement that fails is reported as "no change we can prove", never as zero and never as a trend.
• Trusting a common-word brand name. A mention is a whole-word match of your brand name or domain. A brand called Linear is counted every time an answer says "linear time". The error only ever runs high, so read a few stored answers before you trust the rate.
• Expecting it to predict traffic. Share of voice measures presence in answers. No product can join an answer someone read to a visit they made later.

How do you raise share of voice?

Not by writing about share of voice. The tracked answers name the brands that the pages the engines read already name: roundups, review sites, forums and the brands' own documentation. Who owns the pages the engines cite, counted, is on source citation. The work is to be on those pages and to have your own page answer the question better than the roundup does. That is the work Attensira's crew does: it finds the sources that decide each answer, drafts the fix or the placement for your approval, and reads the next answers to see whether the rate moved by more than the noise.

You can see your own share of voice per engine, with its sample size, on mention rate tracking. The method is documented in full on docs.attensira.com.