● AI search + digital visibility

AI share of voice: what it measures, how to calculate it, and how to read it

AI share of voice is your brand’s share of all the brand mentions in AI answers to a fixed set of buyer questions, counted against the other brands in your category. If the answers name brands 80 times and 12 of those are you, your share of voice is 15%.

Because every brand’s share adds up to 100%, it shows whose name buyers hear when they ask AI tools about your category, not just whether yours comes up. The division is easy. The competitors, the questions and the sample decide whether the number means anything.

Three numbers that all get called share of voice

Teams and tools use the label for different counts, and the same answers can give three different figures. Decide which one you report before you collect anything.

  • Mention share. Your mentions as a share of every brand mention in the answers, counting a brand once per answer. This is share of voice in the advertising sense, and the shares add up to 100%.
  • Answer share. The share of answers that name you at least once. It does not add up to 100%, because one answer can name five brands, and it rises for everyone when answers simply list more of them.
  • Citation share. Your site’s share of the links an answer cites as sources. A brand can be named without its site being cited, and a page can be cited for a point that never names the brand.

The table shows one set of 40 answers read all three ways. The figures are made up.

BrandAnswers naming itCitation shareMention share
Rival A32 of 40, or 80%6%40%
Rival B20 of 40, or 50%14%25%
Your company12 of 40, or 30%9%15%
All other brands16 mentions between themThe rest, split among other sites20%

Your company is at 30% by one count and 15% by another, from the same answers. And Rival A, named most, is cited least, which says its own pages are not what the engines lean on. Report mention share as the headline and keep the other two beside it.

Choose the competitors before you look

The denominator decides the number. Count only the rivals you already track, and every brand the answers name that you left out disappears from the total, which makes your share look bigger than it is.

A study led by Northwestern University, posted in September 2026, makes the same point for research: set the competitive list independently of the AI’s answers, or the brands the models rarely recommend drop out of the analysis by design.

Do both. Fix your competitor list from sales data, the brands you meet in deals and lose to, before the first run. Then add an “others” line that collects every other brand the answers name. If that line keeps growing, a rival you never considered is gaining ground.

Use category questions, not brand questions

A question that names you, such as a comparison with a rival, nearly always gets an answer that names you, so it inflates your share. Keep those for checking what AI says about you, and build share of voice on questions that name nobody.

The wording matters as much. In the Northwestern study, describing a buyer’s needs and constraints changed which brands came back, and brands missing from plain category answers could appear when the question carried the cues they were known for.

So mix plain category questions with ones in your buyers’ words, by role, company size and problem. Keep a fixed list of buyer questions, asked the way buyers see the tool, and change it only when your market changes.

Sample enough answers to trust a change

One run of a question tells you little, because AI answers name different brands from one run to the next. The Northwestern team ran each category question 40 times per model, in separate sessions, before reading anything into the results.

Share of voice is still a sample, so small moves from one month to the next are usually noise. The margin of error on a rate sampled from AI answers is wide at a few runs and narrows slowly as runs are added, so pool questions rather than rerun one.

One thing works in share’s favor. When an engine update makes answers name fewer brands, every brand’s mention count falls, but the shares can hold. That makes share of voice a steadier trend line than raw mentions.

Weight position, but keep the plain number

Being named first is worth more than being fifth in a list. The Northwestern study measured this prominence with mean reciprocal rank, which scores a brand by one over its position: 1 for first, 0.5 for second, 0.2 for fifth and nothing when it is missing.

Use that, or something as simple, and report it beside the unweighted share rather than instead of it. A score that blends position, sentiment and question importance can move for reasons nobody can trace back to an answer.

Report it by engine and by topic

Blend ChatGPT, Gemini, Perplexity and Google’s AI answers into one figure and a gain on one hides a loss on another. Report share per engine, then per topic or buying stage, so each number points at something you can fix.

Only one engine reports a share for you. Bing Webmaster Tools added Citation Share in June 2026: the percentage of all citations shown for a grounding query that went to your site, across Copilot and Bing’s AI answers.

Bing calls it an observational metric, “not a ranking system or a competitive scoreboard”, and it does not show which other domains were cited. Google’s Search Console reports your own impressions in AI Overviews and AI Mode but nobody else’s, so it gives a count, not a share.

What share of voice does not tell you

Share of voice measures who gets named, not who gets chosen. A buyer who meets your name in an answer still has to believe what the answer says about you.

Forrester’s 2027 predictions expect two-thirds of B2B providers to lose deals because buyers cannot verify their claims, as buyers use AI tools to fact check vendors. A share built on claims nobody can check is fragile, which is why the pages AI tools cite need proof, not adjectives.

Northwestern’s team also found that prominence went with search interest and online conversation about a brand more than with conventional brand popularity, an association they do not call cause. Read share of voice as a gauge of your wider visibility, not a lever of its own.

A monthly report that fits on one page

  • Mention share by engine for you, each named rival and “others”, with last month beside it.
  • Answer share for the same brands, so longer lists do not pass for a gain.
  • The position-weighted share, beside the plain one.
  • Bing’s Citation Share for the queries that matter, and your impressions in Google’s AI features.
  • The questions where you gained and lost most, with the answer text, so the numbers lead to a page to fix.

If you change the questions, the engines or the method, say so on the report and start the trend line again. A share of voice series can only be compared with itself.

● About the author

Hamza Hashim is the marketing strategist at Katama. He owns the channel mix and the order it gets built in: what ships first, what waits for evidence, and what gets stopped when the evidence says stop. Meet the team.

Fair questions.

What is a good AI share of voice?

There is no benchmark that travels between categories, because categories differ in how many brands an answer names. Judge it against your rivals and your own trend.

Is AI share of voice the same as share of search?

No. Share of search is your share of the branded searches in a category. AI share of voice is your share of the brand mentions in AI answers. They measure different things, so track both.

Do I need a tool to measure AI share of voice?

Not for a few dozen questions on two or three engines; a spreadsheet works. Tools help at scale, but check how each one sets its denominator before comparing its numbers with yours.

How often should we measure it?

Monthly suits most B2B teams, with the runs spread across the month rather than done on one day, so a single odd day or an engine update does not set the figure.

● Build my growth engine

Tell us where growth is stuck.

We look at where your pipeline comes from today and come back with what we find. No pitch, and if we are not the right fit we will say so.

Rather talk? +1 (774) 777-6454

Goes to hello@katama.io. No sequences, no list. Need to tell us more? Open the contact page.