AI share of voice: how to calculate it, benchmark it and grow it

"AI share of voice" has grown roughly 557% year over year in US search, according to Google Ads data pulled in September 2026. The metric is becoming standard vocabulary for a reason: citation rate alone stops being useful the moment you get cited at all.

A brand that discovers it appears in 40% of tracked answers wants to call the job done. That is premature. The real question is placement: inside that 40%, are you the first brand named with two lines of reasoning, or the fourth item in a list with no justification attached?

How to calculate it without flattering yourself

The simplest calculation counts mentions: across all brands named in your tracked answers, what share is yours. It is a fine starting point, but it treats a passing mention as equal to a strong recommendation, which it plainly is not.

A more honest calculation weights each citation by quality. Three components are enough: position in the answer, because the first brand named gets read and later ones get skimmed; whether a rationale is attached, because a brand described with its arguments converts far better than a bare name; and whether the mention is a recommendation or merely an alternative.

• Define a stable query set that matches what your customers actually ask.
• Measure repeatedly across several days: a single reading captures noise, not a trend.
• Count every brand named, not only the ones you already considered competitors.
• Weight by position and by the presence of a rationale, or you will overstate your standing.
• Track the direction over time rather than the absolute value, which is not comparable across query sets.

The competitive benchmark is the instructive part

The exercise almost always produces a surprise. The brands sitting next to you in AI answers are rarely the ones you bid against. You routinely find smaller brands that are unusually well documented on one specific use case, and marketplaces absorbing a large share of recommendations on generic queries.

That finding is directly actionable. A brand consistently cited beside you for one use case is telling you exactly which attribute your own documentation is missing. It is not a budget problem. It is an absent fact.

Moving the number

Three levers, in order of return. Mechanical blockers first: blocked crawlers, missing product data, an incomplete feed. Then the factual density of the pages tied to the queries where you lose. Then external corroboration, meaning your product facts existing somewhere other than your own site.

In Tadow, share of voice is tracked query by query over time, with the competitors named alongside you and the movement after a fix. The point is not the number itself. The point is being able to say whether what you changed last month moved anything, or did not.