Guide

How to measure AI share of voice

The formula, what counts as a mention and how wide the range is, worked through on 32 real answers from ChatGPT and Gemini, with a calculator for your own counts.

Menzio ResearchChecked November 15, 20268 min read

In short

  • AI share of voice is your brand’s slice of all the brand mentions in AI answers to your buyers’ questions: the answers that name you, divided by the answers that name you or any competitor you track, each brand counted once per answer.
  • Report it with the count behind it. 31 of 142 mentions is 22% (16–29%); 3 of 14 is 21% (8–48%). Nearly the same share, very different certainty.
  • The share depends on which competitors you track, so keep the list fixed, compare like with like, and read your place in the ranking before the percentage.
On this page9 sections

The formula

AI share of voice tells you how much of the conversation about your category goes to you when people ask ChatGPT, Gemini or Google’s AI for a recommendation. Take the answers to your buyers’ questions, count how many name each brand you track, and divide:

Share of voice = answers that name you ÷ (answers that name you + answers that name each tracked competitor)

Most tools publish the same formula: your brand’s mentions divided by all brand mentions [1] [2] [3]. What it leaves open is what counts as one mention, and whose mentions make up the total. As the example below shows, those two choices alone can move the number by several points, so they’re worth spelling out. Semrush, for one, says its enterprise score also weighs where the brand appears in the answer [1]: two tools can report different shares from the same answers.

What counts as a mention

  • Named explicitly. By its name, an alias you set (“Bigin by Zoho” for Zoho CRM) or its domain, written in the answer. “A popular sales CRM” is not a mention.
  • A link is a citation, not a mention. A link, footnote or source that points to the brand’s site is a citation. Citations are counted apart: an answer can cite a brand’s page without naming it.
  • Once per answer. An answer that names a brand five times counts it once, the same as an answer that names it once.
  • Named is named. First place in a list and a passing mention count the same. Where the brand appears is a separate number (position), not part of the share.
  • Tracked brands only. The total adds up your brand and the competitors you chose. Other brands the answers name are worth a list of their own, but they stay out of the total.
  • Real answers only. When a search shows no AI answer, which is common with Google’s AI Overviews, there’s nothing to count.
  • Not your own name in the question. A question like “alternatives to Pipedrive” brings Pipedrive into the answer by design. Questions that name your brand are left out of your share.

How to measure it, step by step

  1. Write the questions your buyers ask, in their words and without your brand name. The AI visibility audit guide covers how to pick them.
  2. Fix the list of brands: yours and the competitors you want to be measured against. Write down the names each one goes by.
  3. Ask each question on each engine more than once. AI answers change from run to run: SparkToro found under a 1 in 100 chance that ChatGPT or Google’s AI gives the same list of brands twice in 100 tries [4]. In our own study, a brand named first came back in the next answer 96% of the time; one named fourth to sixth, 66%.
  4. Mark which tracked brands each answer names, once each, following the rules above.
  5. Add up and divide. Your count over the sum of every tracked brand’s count, kept next to the counts themselves.
  6. Put a range on it before you compare it with last month or another engine (the calculator below does it).

Worked example: 32 answers about CRMs

In our study on how much AI answers change, we asked ChatGPT and Gemini “What is the best CRM for a small marketing agency?” and “Which CRM should a 5-person sales team use?” 8 times each, on November 5 and 6, 2026. That gave 32 answers, every one stored word for word. We took Pipedrive as the brand and the 7 CRMs the answers named most as its competitors, and counted them with the same rules Menzio uses.

Fig. 1

Share of voice of 8 CRMs in 32 AI answers

  • PipedriveYou22% · 31
  • HubSpot22% · 31
  • Zoho CRM20% · 29
  • Salesforce10% · 14
  • monday CRM9% · 12
  • GoHighLevel8% · 11
  • Freshsales6% · 8
  • Attio4% · 6
Source: 32 ChatGPT and Gemini answers to two CRM questions from our study, Nov 5–6, 2026, counted with Menzio’s rules · 142 mentions

Pipedrive was named in 31 answers and HubSpot in 31: a tie for first place, with 21.8% of the 142 mentions each. Zoho CRM came close behind with 29. The other five shared what was left. Notice the two numbers that look alike and aren’t: Pipedrive appeared in 97% of the answers (its visibility), but holds 22% of the mentions (its share of voice), because almost every answer named three or four of the other CRMs on the list too.

The share depends on who you track

Now drop four competitors from the list and keep only HubSpot, Zoho CRM and Salesforce. Same 32 answers, same 31 that name Pipedrive. The share jumps:

Fig. 2

Pipedrive’s share of voice, with 7 or 3 competitors tracked

  • 7 competitors
    22% (16–29%)
  • 3 competitors
    30% (22–39%)
Source: 32 ChatGPT and Gemini answers to two CRM questions from our study, Nov 5–6, 2026, counted with Menzio’s rules · shaded band: 95% range

From 22% to 30%, without a single answer changing. Its visibility stayed at 97% and its place stayed first. That’s why a share of voice is only comparable on the same list of competitors and the same questions, and why the place in the ranking is the steadier number to lead with. When you add a competitor, say so in the report, or the drop will look like lost ground.

Count brands, not repetitions

Counting every time a name appears sounds more precise. On these answers it would have told a different story:

Fig. 3

The same answers, counted two ways

Once per answerEvery repetition
HubSpot22% (31 answers)32% (146 times)
Pipedrive22% (31 answers)27% (122 times)
Zoho CRM20% (29 answers)14% (63 times)
Salesforce10% (14 answers)4% (18 times)
Source: 32 ChatGPT and Gemini answers to two CRM questions from our study, Nov 5–6, 2026, counted with Menzio’s rules · 459 name repetitions

Pipedrive and HubSpot were each named in 31 answers, a tie. Count every repetition and HubSpot moves 5 points ahead, 32% to 27%, without one more answer naming it: the answers that brought it up just said it more often. Salesforce was named in 14 answers, usually just once, and its share falls from 10% to 4%. Repetitions measure how much an answer talks about a product. The buyer reads a list of names, so the question that matters is whether yours is on it.

How sure is the number?

Pipedrive’s 22% rests on 142 mentions. Ask the same questions again next week and it can move a few points without anything changing. A 95% range puts a size on that: the Wilson score interval, a standard way to put a range on a share that behaves well with small counts [5]. For 31 of 142, it gives 22% (16–29%). Try it with your own counts:

Fig. 4

Share of voice calculator, with its 95% range

Answers that name your brand

Answers naming each tracked brand, added up, yours included

Share of voice
22%95% range 16–29%

31 of 142 mentions. The 95% range is 13 points wide; it narrows as the mentions behind it grow.

Wilson score interval, 95%. Starts on the worked example; nothing you type leaves the page

Use the range when you compare. On these answers, Pipedrive held 21% (13–31%) of the mentions on ChatGPT and 23% (14–35%) on Gemini. The ranges overlap almost entirely, so this sample shows no clear difference between the two engines. The same goes for this month against last: if the two ranges overlap, call it no clear change, whatever the headline numbers say.

What the range assumesShow

The Wilson interval treats every mention as a separate draw. Real mentions come in bundles: one answer names several brands, and runs of the same question move together. So read it as a guide to how much your number rests on, not as an exact margin.

It can err either way. Resampling whole answers instead of single mentions gives Pipedrive a narrower range, 21.0%–22.9%, because it is named in 31 of 32 answers and its count barely moves. A brand named in a handful of answers that cluster on one question would get a wider one. With only two questions, as here, the bigger unknown is which questions you chose: more questions do more for it than more runs of the same few.

Share of voice in Menzio

Menzio counts share of voice exactly as above, on ChatGPT, Gemini, Google AI Overviews and Google AI Mode, with your questions asked every day. Its Share of voice page ranks every tracked brand by the answers that name it, with the counts behind each share, your place and how it moved, the trend week by week and your place on each engine. Questions that name your brand are left out. Brands the answers name that you don’t track are listed apart, so you can add one to the list when it keeps coming up.

The 95% range sits on visibility, the share of answers that name you, worked out by resampling whole questions so that runs moving together are accounted for (with fewer than 10 questions or 30 answers, the Wilson interval). For share of voice the page shows the counts, so you can put them in the calculator above. The full rules are on the method page.

Sources

  1. 1.How to measure AI share of voice using Semrush · Semrush · checked Nov 15, 2026
  2. 2.Share of Voice · Peec AI documentation · checked Nov 15, 2026
  3. 3.How to measure your Brand’s Share of Voice on Content AI · Otterly.AI · checked Nov 15, 2026
  4. 4.AIs are highly inconsistent when recommending brands or products · SparkToro · checked Nov 15, 2026
  5. 5.Binomial proportion confidence interval (Wilson score) · Wikipedia · checked Nov 15, 2026

See who AI names in your category

Three buyer questions on ChatGPT, Gemini and Google AI Overviews, free and without signup: which brands get named, and whether you are one of them.

Free · no signup · ChatGPT, Gemini and Google AI Overviews

Keep reading

Changes to this page

  • November 15, 2026: First published.
  • November 15, 2026: A link or source that points to a brand’s site now counts as a citation, not a mention. The worked example is recounted with that rule.