What is AI share of voice, and how do you calculate it?

• 25% vs 40%the same brand's share of voice and visibility in one documented examplePeec AI documentation
• <1 in 100chance that an AI assistant gives the same list of brands twiceSparkToro & Gumshoe, 2,961 runs, 2026
• 0.115weight Ahrefs gives a ChatGPT prompt relative to its Google search volumeAhrefs Brand Radar, October 2026

AI share of voice is the percentage of AI answers that mention your brand, out of the answers that mention any brand you compete with, across a set of prompts your buyers actually ask. If ChatGPT names at least one of your tracked competitors in 240 answers and names you in 66 of them, your share of voice in ChatGPT is 27.5%. That is the whole idea. The hard part is not the division; it is deciding what goes into the top and bottom of it, and knowing how much the result would move if you simply ran the prompts again.

What is AI share of voice?

Share of voice is an old advertising metric: your share of all the advertising, traffic or mentions in your market. Semrush's 2022 guide to it notes that there is no single definition, and gives the search version as your organic traffic divided by the total organic traffic of the market [5].

The AI version applies the same idea to answers from assistants such as ChatGPT, Claude, Perplexity, Gemini and Google's AI Overviews. Ahrefs defines it as the share of AI visibility your brand captures compared with the competitors you track, across the prompts you care about [1]. Semrush describes it as the percentage of mentions your brand receives in AI answers compared with competitors in your market [4].

Three things in those definitions are easy to skip and decide everything:

• It is relative. Your number depends on which competitors you track. Add a strong rival to the list and your share falls without a single answer changing.
• It is bounded by a prompt set. There is no share of voice “in ChatGPT” in general, only across a specific list of questions.
• It counts mentions, not recommendations. A brand named as the tool to avoid still counts as a mention [1].

It is also different from visibility, sometimes called mention rate: the share of all answers that name you, whether or not any competitor appears. Visibility tells you whether you exist in the category. Share of voice tells you how you compare with the brands that do.

How do you calculate AI share of voice?

There are three formulas in common use. They answer slightly different questions, and they give different numbers from the same data.

1. Answer-based: answers that mention you ÷ answers that mention any tracked brand × 100. Ahrefs gives this as the common way to calculate it [1].
2. Mention-based: your mentions ÷ all mentions of tracked brands, including yours, × 100. This is Peec AI's formula [3], and Semrush describes its metric as a share of mentions [4].
3. Volume-weighted: your estimated impressions ÷ the estimated impressions of all answers that mention a tracked brand. This is how Ahrefs Brand Radar weights it [1].

The first two differ in one important way. With the mention-based formula, every brand's share adds up to 100%, like a pie. With the answer-based formula they do not, because one answer usually names several brands: in Ahrefs' own three-prompt example, every brand ends up with the same 67% [1]. Neither is wrong. The pie version is easier to present; the answer-based version is closer to the question a buyer's shortlist raises, which is “how often am I in the answer at all?”.

The volume-weighted version tries to account for the fact that some questions are asked far more often than others. Brand Radar gives each prompt an “AI-adjusted volume”: the monthly Google search volume of the keyword behind it, multiplied by a ratio for each platform. The current ratios are 0.115 for ChatGPT, 0.011 for Gemini and 0.01 for Perplexity, based on how much traffic each platform sends to websites compared with Google [1]. Ahrefs is open that this is a modelling choice: its methodology page says it does not claim a validated link between Google search volume and how often a question is asked inside an AI tool [2].

What does a worked example look like?

The numbers below are invented to show the arithmetic; they are not from a client. Say you run 100 buyer prompts three times each in ChatGPT, which gives 300 answers. You track yourself and two competitors.

Same answers, three defensible numbers for “your share”: 22%, 27.5% and 23.9%. Peec's documentation has its own version of this gap, with a brand at 40% visibility and 25% share of voice [3]. If a report gives you one figure without saying which formula produced it, you cannot compare it with anything, including last quarter's report from a different tool.

Two counting rules also need to be fixed before you start. Count a brand at most once per answer, so that an answer repeating your name five times does not count five times. And decide whether a citation of your website without your brand name counts as a mention; Brand Radar counts a response that mentions the brand or cites its site towards impressions [1].

Why do tools give different numbers?

Ahrefs says it plainly: two tools can report different AI share of voice for the same brand without either being wrong [1]. The differences come from five choices:

1. The formula, as above.
2. The prompts. Brand Radar builds its prompts from Google's People Also Ask data and its keyword database, and says long-tail or niche prompts may be missing [2]. A tool where you write your own prompts measures a different population.
3. The engines and how they are queried. Brand Radar runs prompts through the free public interfaces of each assistant [2]; other tools use APIs. SparkToro's study lists whether API calls match real users' answers as an open question [6].
4. The competitor list. The denominator changes with every brand you add or remove.
5. The time window. Brand Radar refreshes ChatGPT, Perplexity, Gemini and Copilot monthly and reports a 90-day window [2]; a weekly tracker will move more.

My view: for a B2B SaaS company with 20 to 200 people, a market-wide index is a useful outside view, but the prompts that decide your deals are usually too specific to have much Google search volume. I would treat a volume-weighted index as a directional check and base decisions on your own frozen prompt set, built from sales calls and demo requests.

How precise is your number?

Less precise than most dashboards suggest, because the answers themselves are unstable. SparkToro and Gumshoe had 600 volunteers run 12 prompts 2,961 times in ChatGPT, Claude and Google's AI. There was less than a 1 in 100 chance of getting the same list of brands twice, and about 1 in 1,000 of getting the same order. Some answers named two or three brands, and equally often ten or more [6]. What held up was how often a brand appeared across many runs: the authors call visibility across dozens to hundreds of prompts, each run several times, a reasonable metric, and dismiss “ranking position in AI” [6].

So a share-of-voice figure is an estimate with an error bar. A rough 95% margin for a percentage p measured on n independent answers is:

margin ≈ 1.96 × √( p × (1 − p) ÷ n )

The catch is the word independent. Three runs of the same prompt are not three independent observations; they tend to agree with each other. Applied to the example above, 22% visibility gives about ±4.7 points if you treat all 300 answers as independent, and about ±8.1 points if you treat the 100 prompts as the unit. The honest figure lies between the two. On the method page I show how this plays out for different splits of a fixed budget: 200 prompts run three times gives about ±5.4 points, while 20 prompts run 30 times gives about ±15.4.

The practical rule follows from that. If this month's interval overlaps last month's, you cannot say your share changed, however good the headline number looks. More prompts narrow the interval faster than more runs of the same prompts.

How to measure it yourself in five steps

1. Write and freeze the prompt set. 50 to 200 questions in your buyers' words, spread across buying stages: category, comparison, validation. The guide to getting recommended by ChatGPT covers where to find them. Do not edit the set between measurements, or a change in questions will look like a change in visibility.
2. Fix the competitor list. Three to six brands your sales team actually loses deals to. Write the list down with the date.
3. Run each prompt at least three times per engine in the same window, and record every brand named and every URL cited.
4. Calculate per engine, not blended. Report visibility and share of voice for each engine separately, with the formula named and a margin of error. A blended score hides which engine moved, so you cannot tell what to fix.
5. Re-run on a schedule and compare like with like. Same prompts, same competitors, same engines. Monthly is usually enough: answers vary from run to run, and changes to third-party pages take weeks to filter through.

Keep the list of cited URLs from step 3. It is the most useful by-product of the whole exercise, because it tells you which third-party pages shape the answers, and that is usually where the work is.

What is a good AI share of voice?

There is no universal benchmark. Ahrefs notes that benchmarks are only comparable when the methodology matches [1], and with three formulas, different prompt sets and different competitor lists, a figure such as “30% is good” has nothing to stand on.

A more useful comparison is with your own market share. Advertising research has long held that brands whose share of voice exceeds their share of market tend to grow; LinkedIn's B2B Institute, citing work with Les Binet and Peter Field, puts it at about 0.7% of annual market share growth for every 10 points of excess share of voice in B2B [7]. That finding is about advertising spend, not AI answers, and nobody has shown that it carries over. My view: it is still a sensible frame. If you hold 15% of your market but appear in 5% of the answers your buyers read, the gap is the problem worth explaining, whatever the absolute number is.

What does share of voice not tell you?

• Whether the mention helps you. Share of voice does not show whether mentions are positive, or whether the assistant recommends you [1]. Read a sample of the answers.
• Traffic or revenue. Ahrefs describes its impressions as a modelled estimate of exposure, not a count of people who saw an answer [1], and calls Brand Radar's metrics directional indicators rather than traffic counts [2].
• Why you are missing. The cited sources do that. Share of voice tells you where you stand; the citation list tells you what to change.
• Position in the answer. I do not weight share of voice by where a brand appears in the list. Given how often the order changes between runs [6], position adds noise rather than information.

My view: share of voice is worth tracking, but only as one line in a report that also shows visibility, citations and the margin of error. A single blended number with no interval is the AI-search version of a vanity metric.

Questions

What is the formula for AI share of voice?

The most common version is answers that mention your brand divided by answers that mention any tracked brand, multiplied by 100. Some tools divide your mentions by all tracked-brand mentions instead, and Ahrefs Brand Radar weights each prompt by estimated search volume. Always check which one a report uses.

What is the difference between AI visibility and AI share of voice?

Visibility is the share of all answers that mention you, with or without competitors. Share of voice compares you only with the brands you track. The same brand can show 40% visibility and 25% share of voice, as in an example in Peec AI's documentation.

How many prompts do I need to measure AI share of voice?

Enough that the margin of error is smaller than the change you want to detect. In practice that means at least 50 prompts and usually 150 to 200, each run three or more times per engine. More distinct prompts narrow the interval faster than more runs of the same prompt.

Can I compare AI share of voice between two tools?

Not directly. Tools differ in formula, prompt source, engines, competitor list and time window, and Ahrefs says two tools can report different figures for the same brand without either being wrong. Compare a tool only with itself, on the same settings.

How often should I measure AI share of voice?

Monthly is usually enough for a B2B SaaS company. Answers vary from run to run, and most real changes, such as being added to third-party comparison pages, take more than a few weeks to show up in a measured share.

Want your share of voice measured properly?

The AI visibility audit runs your buyers' prompts across ChatGPT, Claude, Perplexity and Google AI Overviews, lists the pages that get cited, and turns it into a prioritised backlog. The scope is free and comes within one working day.

Get a free scope →

Sources

Checked on 9 November 2026. Figures are quoted as published; samples and periods differ between studies.

1. Ahrefs — AI share of voice: what it is, how to measure it, and what it actually tells you, by Si Quan Ong (22 October 2026)
2. Ahrefs — Brand Radar methodology: how we collect and model AI visibility data (29 October 2025, updated 26 February 2026)
3. Peec AI documentation — Share of Voice (undated, checked 9 November 2026)
4. Semrush Knowledge Base — AI visibility metrics (undated, checked 9 November 2026)
5. Semrush — Share of voice: what is it and how can you increase yours?, by Brian Moseley (20 June 2022)
6. SparkToro — AIs are highly inconsistent when recommending brands or products, 2,961 runs (27 January 2026)
7. LinkedIn B2B Institute — The B2B marketer's guide to the “share of voice rule” (17 January 2020)