AI Share of Voice

A rate and a share answer different questions. A brand cited in half of its tracked answers can be gaining ground or losing it: if competitors are named more often in the same answers, its share falls while its own rate, the AI search visibility figure, holds still. Share of voice puts the competitors in the denominator, which is how buyers meet an answer: as a shortlist. Microsoft’s advice to marketers makes the same point from a vendor’s side: in AI search, visibility is no longer defined by ranking position alone but by which pieces of content earn a place in the final answer^[3].

A vendor metric of this kind exists: Bing’s Citation Share, in preview in Bing Webmaster Tools since June 2026: the percentage of citations attributed to a site out of all citations shown across all sites for the same grounding query^[1]. Microsoft is careful about its limits. It calls Citation Share an observational metric, not a ranking or a competitive scoreboard, and says it exposes no competitor domains and represents no traffic share^[1].

Shares move for reasons outside the page. Microsoft lists changes in user behaviour, evolving models, freshness signals and partner refresh cycles^[1], and each AI surface answers with its own models and sources. A share is therefore read per surface and as a trend, never as one number across all of them.

Segments change the picture as well. In QueryRev’s study, places 1 to 1,000 of the panel block an AI crawler by name at 17.6% of the sites read, against 13.9% for places 1,001 to 10,000^[4]: one figure for the whole panel would hide the difference. A share of voice across all prompts hides which topics a competitor owns in the same way.

From our practice: in September 2026 an e-commerce furniture brand in the US saw its AI visibility score fall from 31 to 18 while its citations rose to 76, and the client report put the fall down to off-topic prompts dropping out of the tracked set. A share is comparable across periods only on the same prompts.

Fix the prompt set, the AI surfaces and the competitors first; a share is defined only for that set. For each answer, record every tracked brand that is mentioned and every one whose page is cited. Mention share is a brand’s mentions over the mentions of all tracked brands; citation share is the same for citations. Report the two apart.

Choose the counting unit and say it. Counting each brand once per answer measures presence; counting every citation measures weight, and a brand cited three times in one answer scores differently under the two. Bing counts citations per grounding query and across all sites^[1]; a share from tracked prompts counts per prompt, among the brands you chose. Bing’s AI Performance report beside it gives the citation counts per page^[2].

Split before you sum: by surface, by topic and by intent. Weight buying prompts above definitions if the business does, and state the weights beside the figure.

Sample data: an invented site, to show the method.

A bed retailer tracks 40 buying prompts on two AI surfaces against three competitors: 80 answers. Across them the four brands are mentioned 110 times. The retailer has 34 of the mentions, 31%; the leader has 41, 37%.

Split by topic, the gap sits in one place: on prompts about mattresses for back pain, the leader is named in almost every answer and the retailer in few. Its own mention rate looked healthy; the share showed whom the answers favoured.