A brand’s share of measured appearances and citations relative to competitors within an AI response dataset.
AI share of voice compares a brand’s measured presence with that of a defined competitor set in an AI response dataset. There is no single counting method implied by the name. A report might use brand mentions, recommendations or citations, so the unit must be stated before the percentage can be interpreted.
One transparent approach counts each brand at most once per eligible response. The brand’s count is divided by the sum of those brand-response counts for the tracked set, then multiplied by 100. A response may count for several brands under this method. This is a share of tracked appearances, not the percentage of responses containing your brand.
The competitor list and sampling method affect the result. Adding a competitor, changing prompts or weighting some responses more heavily can alter the share. Disclose the platform, dates, prompt sample, identity-matching rules and any weights. When the total measured appearances are zero, the share is undefined.
Takeaway: Choose a method that answers your comparison question and keep it stable. Do not mix citations and recommendations into an unexplained competitive score.
Example
Across a defined test, Brand A appears in 12 eligible responses and the tracked brands collectively accumulate 60 brand-response appearances. A has 20% of those appearances. If the test had 40 responses, A’s response appearance rate would instead be 30%; the two figures describe different things.
What to check
Publish the numerator, denominator and competitor set with the result. Preserve ambiguous matches for review. A sampled share of voice is not a market-share, revenue-share or platform-wide endorsement measurement.
For more detail, see AI features and your website.