What Is AI Share of Voice? A Practical Calculation Guide
AI share of voice measures a brand’s relative presence among tracked brands within a selected answer sample. It is not sales share, market share or a universal visibility percentage. The result is only interpretable when the brand list, question set and counting rule are explicit.
The SI visibility guide treats it separately from mention rate and source citations. These metrics can tell different stories about the same answers.
Why do mention rate and share of voice use different denominators?
Mention rate asks how many answers contain the brand. Share of voice asks how much of the tracked brand presence belongs to it. They need not be equal.
| Metric | Numerator | Denominator |
|---|---|---|
| Mention rate | Valid answers containing the brand | All valid answers |
| Answer-based share of voice | Brand’s answer-level mentions | Answer-level mentions of all tracked brands |
| Website citation rate | Answers citing the website | All valid answers |
Here, a brand counts once per answer even if its name appears five times. Other tools may count repetitions or weight positions. Do not compare these methods as if they were identical.
An illustrative calculation
This is invented example data, not a customer result. Suppose 20 valid answers include three tracked brands:
| Brand | Answers containing it | Mention rate | Share of voice |
|---|---|---|---|
| A | 8 | 8 / 20 = 40% | 8 / 20 = 40% |
| B | 7 | 7 / 20 = 35% | 7 / 20 = 35% |
| C | 5 | 5 / 20 = 25% | 5 / 20 = 25% |
The total brand mentions happen to equal the answer count. If C appeared in 10 answers instead, there would be 25 brand mentions. A’s mention rate would remain 40%, while its share of voice would become 8 / 25 = 32%. Its relative share falls even though its own presence stays constant.
Define the competitor set
Adding a competitor can increase the denominator. A lower percentage may therefore reflect a wider comparison group rather than a loss of mentions. Show the brand list and change dates. Also record brands appearing outside the tracking list rather than silently ignoring them.
Map genuine naming variants together without merging different companies. Decide whether a product and its parent company count as one entity. Check several raw answers manually before applying the rule to the full dataset.
Review context and accuracy
A frequently mentioned brand may be described with an obsolete price, unsuitable positioning or negative context. Review factual accuracy, recommendation context and citations alongside the percentage. Qualified enquiries on the website provide another layer.
Report language, platform and decision groups separately. In small samples, one changed answer can shift percentages substantially. Keep prompt coverage stable and repeat measurements before declaring a competitive win.
The formula on this page is an explicitly defined working method, not a universal ranking score supplied by an AI provider. Google’s AI search documentation also explains that responses and sources can vary. Continue with AI visibility measurement or the optimization service.