How to measure share of voice in AI answers

What is share of voice in AI answers?

Share of voice is one brand's mentions divided by the mentions of your own brand and every competitor you monitor. A mention is the brand's name in the text of an answer. A brand counts once in an answer, however often the answer repeats its name. A name that appears only in a source link beside a sentence is not a mention. The spellings that FullReach AI knows for a brand count as the brand, and you can add more.

share of voice = mentions of the brand ÷ mentions of your brand and every monitored competitor

Share of voice asks how many of the mentions of your brand and your monitored competitors go to one brand. Brands that you do not monitor do not count at all. Every mention counts the same, wherever it appears in the answer and whatever it says. A warning against a brand counts as much as a recommendation. Sentiment and recommendation status measure that difference. Share of voice in search tools often weights a result by its rank. This share of voice does not.

Visibility is the other headline number: the share of all answers that name the brand. Visibility asks how often a buyer sees the brand at all. The two percentages use different bases, so a visibility of 19% and a share of voice of 35% are not sizes that you can compare.

Media share of voice counts coverage across publications. AI share of voice counts answers to the questions that you chose. The two numbers use different bases, so do not add them or compare them point for point.

What did we measure?

We used our own project, which tracks the category we sell in: AI visibility tools. It asks 48 prompts every day in Ireland, in English. It asks them on 7 AI platforms: ChatGPT, Gemini, Perplexity, Copilot, Google AI Mode, Google AI Overviews and Grok. No prompt names our own brand. Over three days, from 25 to 27 September 2026, it collected 1,008 answers, the same number on each platform.

The table shows every monitored rival that the answers named. Brands A to E are companies in the category. The project tracks our own brand too. It adds no mentions to these shares, so the table leaves it out. Each number pools the answers from all platforms into one ratio.

The answers come from 48 questions asked on three days, so they are not 1,008 separate samples. This is one category in one market over three days. Read the numbers as an example, not a law.

Why was share of voice higher than visibility here?

There are two reasons. First, 72% of the answers named no monitored brand. Those answers change no brand's share of voice, but they lower every brand's visibility. Second, an answer that named any monitored brand named about two of them on average. So the mentions split between rivals.

Share of voice is higher than visibility when the monitored brands together get fewer mentions than there are answers. In our data, they got about one mention for every two answers. In a crowded category, where most answers name several monitored brands, share of voice can be lower than visibility.

The result: a brand can hold a third of the share of voice and still appear in only one answer in five. Read the two numbers together. Share of voice tells you how you compare with your rivals. Visibility tells you how often a buyer sees you at all. The FullReach AI share of voice tile shows, under the share, how many answers name any monitored brand.

Why does share of voice change when no answer changes?

The denominator is the list of brands you monitor. Add a rival that the answers name, and every other brand's share falls. Remove one, and every other share rises. A rival that no answer names changes nothing.

In our data, Brand B held 27% of the share of voice. When we took Brand A, the leader, off the list, Brand B's share rose to 42% on the exact counts. Not one answer changed.

FullReach AI counts every period again with the current list. After you change the list, the dashboard, the weekly digest and the monthly report show past weeks with the new list too. So they compare like with like. A report that you already sent or exported keeps the old numbers, so the two no longer match. Write the list of monitored brands beside every share of voice that you report. FullReach AI asks before it stops monitoring a rival, because every share of voice is then calculated without that rival.

Why does share of voice differ between AI platforms?

Some platforms name more brands in one answer than others. In our data, when an answer named any monitored brand, Grok named 2.9 of them on average, and Gemini named 1.6.

A crowded answer splits the share between more brands. It also adds more mentions to a pooled share of voice, so a crowded platform weighs more in the total. On Gemini, Brand A appeared in 14% of the answers and held 38% of the share of voice. On Grok, it appeared in 28% of the answers and held only 30%. On ChatGPT, it appeared in 30% of the answers and held 38%. On these three platforms, only 20 to 45 answers named Brand A, so treat the platform figures as an illustration. Compare platforms on visibility first, and on share of voice second.

How many prompts does a stable share of voice need?

Most answers name no monitored brand, so the share rests on the few answers that do. We drew 3,000 random prompt sets of each size from the 48 real prompts, with replacement. The table shows where 9 of 10 sets put Brand A.

In percentage points, share of voice varies more than visibility with few prompts, and about as much with 50. Compared with its own level, visibility varies more. A set whose answers named no monitored brand has no share of voice, so the share-of-voice ranges leave such sets out.

So even our own figure of 35% is an estimate, with a 90% range of about 30% to 41%. That range covers the choice of prompts only, not day-to-day variation or a new model version. Our guide on how many prompts to track explains the method.

Why did a share of voice jump?

Check these causes in this order:

1. The list of monitored brands changed. Somebody added or removed a rival, or added a spelling of a brand's name.
2. The view changed. A filter now shows another platform, market or date range.
3. The prompts changed. Somebody added, paused or reworded prompts.
4. The answers changed. You or a rival appear in more or fewer answers than before. Check visibility for every brand.
5. The window is small. A week of few prompts rests on few answers.

Causes 1 to 3 move the number without changing any answer. Causes 4 and 5 are changes in the answers. Cause 4 can come from the market or from a new model version, and cause 5 is chance. FullReach AI records the model version behind every answer. Its weekly digest lists a brand whose share of voice moved by 10 percentage points or more.

To see which side moved, compare each brand's visibility, which is its mentions per answer. With the same number of answers, you can compare the mention counts. Say that your brand had 30 mentions and your rivals had 70: a share of 30%. The next month has as many answers. You have 25 mentions and your rivals have 45. Your share rises to 36%, but you lost 5 mentions.

How do you use share of voice well?

• Monitor the rivals your buyers compare you with. Every monitored name that the answers mention takes part of the share. A long list of small names makes every share smaller.
• Keep the list stable, and write it down. State the monitored brands beside every share you report.
• Read it next to visibility. When your share rises and your visibility falls, you appear in fewer answers than before. Your rivals only fell further.
• Leave out questions that name your brand. An answer to a question that names you almost always names you too. FullReach AI leaves those branded prompts out of every rate by default.
• Filter it by platform and market. An average across platforms mixes answers that name one brand with answers that name five.

For how FullReach AI collects the answers, read the methodology.