AI Share of Voice: Formula, Free Calculator and How to Read It

What is share of voice in AI assistants?

AI share of voice is the share your brand takes in the recommendations of assistants like ChatGPT, Gemini or Grok, compared with your competitors, across a fixed list of questions. It answers one simple question: when an assistant names brands in your industry, what fraction of those mentions goes to you?

Take a recruiting agency in Lille. The founder asks ChatGPT “which recruiting firm for sales profiles in northern France?”. Two competitors come up, not him. He tries again the next day: one competitor, plus his agency in third place. What is that result worth? Nothing, until it is counted across several questions, several assistants and several runs. That is what share of voice is for: turning answers that change with every call into a number you can track.

What you count are the answers that name you unprompted, without your name in the question. A question that already contains your brand measures nothing.

What is the exact formula for AI share of voice?

Share of voice is your mentions divided by the sum of your mentions and those of the competitors you track, multiplied by 100. In this calculation, a mention is an answer that cites the brand: an answer that names you three times counts once.

Share of voice = brand mentions ÷ (brand mentions + tracked competitor mentions) × 100
Presence       = answers citing the brand ÷ answers analyzed × 100

This is the definition Tonaily applies in its AI visibility tracking: brand mentions divided by the total for the brand plus tracked competitors, on the latest run of the panel. Two rules follow from it. With no competitor tracked, share of voice has no comparative denominator: it is 100% by construction, so Tonaily does not display it. And the brand and its competitors must come from the same run, otherwise you are comparing two snapshots from different moments.

Not every definition uses the same denominator, and that is the first source of confusion between tools:

Before comparing two share of voice numbers, always ask which denominator they rest on. A 40% against two chosen competitors and a 15% against every brand in an industry can describe exactly the same situation.

What is the difference between share of voice, presence and rank?

Presence tells you how many answers cite you, share of voice tells you what fraction of the mentions goes to you, and rank tells you whether you are named first or last in the list. These are three different measures, and a brand can be strong on one and weak on another.

Presence is the most robust measure, because it does not depend on the list of tracked competitors. Share of voice, on the other hand, drops if you add a strong competitor to your tracking, even though nothing changed on your side. That is why Tonaily’s AI visibility tab puts presence front and center (an “of answers cite you” gauge, based on the number of answers collected), while share of voice sits on the dashboard.

Rank matters too: being cited first does not carry the same weight as being cited fifth. Tonaily’s AI Visibility Index measures it with MRR (mean reciprocal rank): for each answer, 1 if the brand is cited first, 1/2 if second, 1/3 if third, 0 if absent, then the average across all answers. Each engine’s score combines both, and the published score averages the engines:

Score = 100 × (0.70 × coverage + 0.30 × MRR)

Coverage is presence (the share of answers that cite the brand). It weighs 70%: being cited often beats being cited once at the top. Academic research makes the same choice to weight position: the paper that coined GEO measures a source’s visibility with a “Position-Adjusted Word Count”, which decreases the further down the answer the citation appears (Aggarwal et al., 2024).

The August 2026 edition of the Index for CRM software (30 questions in English, asked to OpenAI’s gpt-5.4-mini, Gemini 3 Flash preview and Claude Haiku 4.5, for 90 answers) shows why the three measures need to be read together:

HubSpot is cited in every answer, yet its share of voice stays under 20%: each answer names a little over five brands on average, so nobody can top roughly 20% on this denominator. Salesforce comes first far more often than Pipedrive, but Pipedrive shows up in more answers and beats it on score. Share of voice alone would have shown neither.

How do you calculate your AI share of voice, step by step?

You need a fixed panel of questions without your name, several assistants, a competitor list set in advance, and a count of the brands cited in each answer. The calculation itself is a single division.

1. Lock the panel. 10 to 20 questions a prospect would ask without knowing your brand or your competitors’. The method is laid out in the guide choosing which questions to track in AI assistants.
2. Pick your competitors before you measure. The ones that actually come up in the answers, not just the ones you run into in sales meetings.
3. Ask every question to every assistant. 10 questions across 3 assistants give you 30 answers.
4. Count, answer by answer, which brands are cited. A brand named twice in the same answer counts once.
5. Calculate presence and share of voice for each brand, with the calculator below.
6. Repeat at the same pace, with the same questions, and compare runs with each other.

Fictional example. An accounting firm tracks 10 questions across 3 assistants, or 30 answers, against two competitors. Its name appears in 6 answers, competitor A in 14, competitor B in 9 (these are the calculator’s prefilled values). Total mentions: 29.

Now add a third competitor, cited in 16 answers. The total rises to 45 mentions and the firm’s share of voice falls to 13%. Its presence stays at 20%: it lost nothing, the rules of the game changed. Conversely, if competitor A drops from 14 to 8 mentions, the firm’s share of voice climbs to 26% without it being cited a single extra time.

Why does your share of voice move when nothing changed?

Because it rests on a small sample of non-deterministic answers: a few mentions more or less are enough to shift it by several points. Before you call a rise or a drop, rule out noise and changes in the measurement.

Small samples. On 30 answers, one mention is worth just over 3 points of presence. The most common approximation of the confidence interval for a proportion, described in the NIST statistics handbook, puts the 95% margin for a 20% presence on 30 answers at about 14 points: the interval runs from roughly 6% to 34%. With 60 answers (20 questions across 3 assistants), the margin drops to about 10 points. NIST notes that this approximation can be inaccurate when the sample is very small, and the answers from one panel are not independent: treat these margins as an order of magnitude, not a guarantee.

Model variability. The same question asked twice can produce two different lists of brands. On benchmark tasks, researchers measured accuracy variations of up to 15% between identical runs of models configured to be deterministic (Atil et al., 2024). This is a documented fact, not a glitch in your tool.

Measurement changes. A rewritten question, an added competitor, an assistant outage that returned 20 answers instead of 30: each one moves the number without saying anything about your real visibility.

flowchart TD
  A[Share of voice moved] --> B{Panel changed?}
  B -- Yes --> C[Compare from the change on]
  B -- No --> D{Competitors changed?}
  D -- Yes --> E[Read presence instead]
  D -- No --> F{Fewer answers?}
  F -- Yes --> G[Wait for a full run]
  F -- No --> H{Presence moving too?}
  H -- No --> I[A competitor moved]
  H -- Yes --> J{Trend over 3 runs?}
  J -- No --> K[Probably noise]
  J -- Yes --> L[Real change]

What share of voice should you aim for?

There is no universal good share of voice: it depends on how many competitors you track and on your industry. The simplest benchmark is the equal share, 100 divided by the number of brands being compared.

With one tracked competitor, the equal share is 50%. With two, 33%. With three, 25%. Above it, you are cited more than your group’s average; below it, less. On the “every brand cited” denominator, the ceilings are much lower: in the CRM example, the leader cited everywhere stays under 20%.

Three comparisons make sense:

• Against the equal share, to know whether you are above or below the pack.
• Against the leading competitor, to measure the gap to close, question by question.
• Against yourself, on the same panel and the same competitor list: it is the only comparison that tells you whether your efforts are paying off.

The “average” numbers published by tool vendors each rest on their own questions, assistants and denominators. They are not comparable to yours.

Choosing competitors is therefore a measurement decision, not a detail. Tracking a giant that assistants cite everywhere crushes your share of voice without teaching you anything; tracking a small player that is never cited inflates it artificially. Track the ones that actually come up in the answers to your questions.

How do you grow your share of voice in AI assistants?

By winning mentions on the questions where a competitor shows up instead of you: a page that answers the question precisely, mentions on the sites assistants read, and a site their crawlers can access. Nobody directly controls what an assistant answers; what follows raises your odds of being cited.

1. Start from the questions you lose. Find the questions where a competitor is cited and you are not, then read the answers: which criteria the assistant highlights (price, company size, industry, region), which words it uses. The guide why ChatGPT recommends your competitor helps you find the cause.
2. Look at the cited sources. When an assistant searches the web, it relies on specific pages. OpenAI documents that, by default, responses from its API with web search include citations of the URLs found, and that a field lists every page consulted (OpenAI, Web search). A comparison page or a directory cited across several of your questions is a page where earning a mention is often worth more than one more post on your blog.
3. Publish the page that answers. One page per use case or customer profile, stating clearly who your offer is for, at what price and with what differences. The principles are laid out in the guide getting cited by ChatGPT, Gemini and Grok.
4. Make your site readable. An AI search crawler blocked in robots.txt or by a firewall cannot read your pages. Check it with the guide is your site blocking ChatGPT?.
5. Provide verifiable evidence. In the GEO study, adding source citations, quotations and statistics improved the visibility of the tested pages, while keyword stuffing brought little or no improvement; the authors also note that effectiveness varies by domain (Aggarwal et al., 2024). This is a research result on a benchmark, not a rule published by the assistants.

On Google’s side, the official position is restrained: there are no additional requirements or special optimizations to appear in AI Overviews or AI Mode, and SEO best practices still apply (Google Search Central).

Then measure the effect on the same panel. New content takes time to be discovered and then cited: compare several runs before and after, and track presence on the targeted questions first, before overall share of voice.

What are the limits of this measure?

Share of voice measures what certain assistants say, on certain questions, at a given moment: not what each of your prospects sees. Read it as a barometer, not as an audience figure.

Answers can vary with language, country, the user’s history and the assistant’s mode (with or without web search). A measurement taken through the API can therefore differ from what a user of the consumer app sees. An assistant that cites many brands per answer also mechanically changes everyone’s share: that is why the AI Visibility Index averages scores per engine rather than adding up raw mentions.

Finally, no tool covers everything. Tonaily’s paid plans track ChatGPT, Gemini and Grok, not Perplexity, Claude or AI Overviews. The free visibility test, which needs no account, asks one question to ChatGPT, Claude, Gemini and Perplexity: useful for a first look, but a single question does not give you a share of voice. To compare tracking tools, their engines and their prices, see the guide AI visibility tracking tools compared.