Definitions · September 28, 2026 · Updated October 1, 2026 · 7 min read
AI Share of Voice: How to Measure It and What It Means
AI share of voice is the share of AI answers naming any tracked brand that name yours. Here is how it is computed, how to read it, and where it misleads.
Share of voice is the number most teams end up watching weekly, and it is the one most likely to be misread. In AI answers it means the share of sampled answers that name you, out of the answers that name any tracked brand (you or a competitor you track), across a fixed set of category questions: if 400 answers from last week’s runs named at least one tracked brand and 100 of them named you, your share of voice is 25 percent. The metric started in advertising, moved to search, and now describes who the assistants recommend.
This post shows how to compute it, how to read it next to the other metrics, and the three ways it lies. The one-paragraph definition lives in the glossary.
The definition, precisely
Start with a prompt set: 15 to 25 questions phrased the way buyers ask them. Run each prompt many times across the assistants you track. For every answer, list the tracked brands it names: you and the competitors you track.
- Visibility rate is the share of answers that name you at all.
- Share of voice is the answers that name you, divided by the answers that name at least one tracked brand.
- Average rank is your mean position among the tracked brands named when you are named.
- First-mention rate is the share of answers where you are the first tracked brand named.
Share of voice counts presence per answer, not raw mention counts. An answer that names you three times counts once. One answer can name several brands, so the shares of all brands can add up to more than 100 percent. Some tools count raw mentions instead, so the same answers can give different numbers in different tools.
Share of voice is the one that puts you and your competitors on the same scale. The rest describe you alone. That is why it is the headline number, and why it needs the others next to it. We define all four in what AI visibility is.
A worked example
Take one prompt, “best help desk for a two-person startup”, asked 20 times across two engines over a week, 40 answers total. Suppose 30 of the 40 answers name at least one tracked brand, and the other 10 name none of them.
| Brand | Answers naming it | Share of voice |
|---|---|---|
| Rival A | 27 | 90% |
| You | 24 | 80% |
| Rival B | 15 | 50% |
| Rival C | 6 | 20% |
Your share of voice on this prompt is 80 percent: 24 of the 30 answers that named a tracked brand. The four shares add up to 240 percent, because most answers name more than one brand. Your visibility rate is 60 percent (24 of all 40 answers). Share of voice is never lower than visibility rate, because its denominator leaves out the answers that name no tracked brand. Your average rank might be 2.3, and your first-mention rate might be 15 percent. Four numbers, one story: you are usually in the answer, rarely first, and Rival A owns this question.
The example is illustrative. Real numbers come from your own runs, which is the point of having them.
Three ways share of voice misleads
1. It is relative, so it can rise while you fall. Suppose some answers named Rival B and no other tracked brand, and a model update stops naming Rival B there. Those answers leave the denominator, so your share of voice goes up with no change in how often you are named. Always read share of voice next to visibility rate. Share up and visibility flat means a competitor lost, not that you won.
2. Small samples make big swings. Five answers cannot support a percentage. Ten is a floor, and the trend over weeks matters more than any day. CitePrism shows the sample size next to each rate and greys out headline rates below ten answers for this reason; whatever you use, insist on seeing the sample size.
3. It depends entirely on the prompt set. Change the questions and you change the number. A prompt set weighted toward enterprise questions will show a different leader than one weighted toward small-team questions, and both are correct for what they measure. Keep the set fixed while you compare periods, and when you change it, mark the date on the timeline so the discontinuity is not mistaken for a trend.
Per engine, not just overall
The engines disagree more than most people expect. It is common to lead share of voice on Perplexity, trail on ChatGPT and be absent on Google AI Overviews, because each one retrieves from different sources and holds a different prior about your category. An overall number hides this. Read share of voice per engine, and treat the disagreement as information: it tells you where to focus, and which sources to look at first.
LLM share of voice tracking in CitePrism computes the metric per engine and overall, with competitors scored by the same prompts, on the same schedule, by the same detection rules, so the comparison holds.
What to do with the number
Share of voice tells you who the assistants prefer. It does not tell you why. The why is in two places:
- The gaps. The specific prompts where a rival is named and you are not. Fix those one at a time rather than trying to move the average.
- The citations. The sources the assistants cite when they answer those prompts. Citation tracking groups them by category so you can see whether the fix is a review site, a community, a comparison article or your own site.
A rising share of voice with no change in the citations you appear in is probably a model update. A rising share of voice after your new comparison page starts being cited is probably you. Keeping both series side by side is how you tell them apart, and it is why we recommend improving AI visibility with the measurement running the whole time.
FAQ
How is AI share of voice different from search share of voice?
Search share of voice estimates the fraction of clicks or impressions a site earns for a keyword set, using ranking positions and click curves. AI share of voice counts which brands each generated answer names, where there are no positions in the search sense and no click data. The name is shared; the arithmetic and the data source are not.
What is a good AI share of voice?
There is no benchmark that transfers between categories. Shares can add up to more than 100 percent, so there is no fixed fair share: it depends on how many tracked brands a typical answer names. Compare against your own history and against the rivals in your prompt set, not against a number from another industry.
Should share of voice include brands I do not track?
In CitePrism it does not. The denominator is the answers that name at least one tracked brand: you or a competitor you added. An answer that names only untracked brands does not count, so a missing rival makes every share read higher. Add the competitors buyers really compare you with. Other tools use other rules, such as raw mention counts or an “other” bucket, so check which rule a tool uses before you compare numbers across tools.
How many runs do I need before trusting share of voice?
At least ten answers per prompt and engine before reading a percentage, and a few weeks of daily runs before reading a trend. Fewer than that and day-to-day movement is mostly the randomness of the assistants, not a change in your position.