AI share of voice: what it means and how to measure it

AI share of voice meaning, in plain words

AI share of voice is your slice of the conversation when AI assistants talk about your category. It counts how often your brand is mentioned in answers, as a share of how often all the tracked brands are mentioned.

Peec AI's documentation gives the formula plainly: your brand's mentions divided by the total number of brand mentions, times 100. Its own example has a brand mentioned 4 times and a competitor mentioned 12 times, which gives the brand a share of voice of 25 per cent.

Semrush's guide defines it as how visible your brand is compared to competitors in your market. The manual formula it gives is the same idea: your AI mentions divided by total AI mentions across all brands in your category.

The key word is share. Share of voice does not tell you how often you are named. It tells you how your mentions compare with everyone else's.

Where share of voice comes from

Share of voice began as an advertising measure of spending, not of mentions. It meant your advertising budget as a share of all advertising in your category.

Wikipedia's entry describes it as the percentage of media spending by a company compared to the total media spending for the product, service or category. Its example is a company spending $5 million in a market where $100 million is spent in total, which is a 5 per cent share of voice.

The AI version keeps the shape and swaps the thing counted. Instead of dollars spent on ads, it counts mentions earned in answers. That swap matters, because nobody can buy a mention in a ChatGPT answer the way they buy airtime. AI share of voice is closer to an earned measure, like press coverage, than to a media budget.

Three ways tools count AI share of voice

Tools agree on the idea and differ on the arithmetic. The differences decide whether two share of voice figures can be compared at all.

The first two differ in a small but real way. In Profound's wording, the top of the fraction counts responses that mention you, so an answer that names you three times counts once. Peec AI's wording counts the times your brand was mentioned. Neither page read says exactly how a name repeated within one answer is handled, so ask before you compare the two.

The weighted method needs the most care. Ahrefs' methodology page says that weighting mentions by Google search volume is a modelling choice. It also says Ahrefs does not claim a validated link between Google search volume and how often a query is asked inside an AI tool. A weighted share of voice can be useful, but it rests on an assumption the vendor itself flags.

Position weighting rests on something shakier still. Where a brand appears in an answer changes from one run to the next, as the section on noise below shows. A figure that rewards being named first inherits that instability.

Share of voice is not the same as visibility

Visibility counts the answers that name you. Share of voice counts your mentions against everyone else's. The two can move in opposite directions.

Peec AI's own example shows the gap. A brand named in 4 of 10 chats has a visibility of 40 per cent. If a competitor is mentioned 12 times in the same chats, the brand's share of voice is 25 per cent. Same answers, two different numbers, both correct.

The next table is a made-up example of how share of voice can rise while nothing improves for you. The brand and both months are invented, and the figures are illustrative only.

In month 2 you were named in exactly as many answers as before. A report that showed only share of voice would call it a good month. The visibility row says nothing changed for you.

Share of voice tells you where you stand against the rivals you track. Visibility tells you whether buyers find you. Report both.

Mentions, citations and which one share of voice counts

Most AI share of voice figures count mentions, not citations. A mention is your brand named in the text of an answer. A citation is a link from the answer to a page, usually shown as a source.

The two often disagree. Ahrefs' methodology page describes a category it calls found but not cited, for brands that appear in an answer without a link to their own pages. An assistant can recommend you by name while linking only to a review site, or link to your page while recommending a rival.

So ask which one a share of voice figure counts before comparing it with anything. A share of mentions and a share of citations are different measures with the same label. Our page on AI citation tracking explains how a link in an answer is matched back to the site that owns it.

Why to count each assistant apart

A single share of voice across all assistants can hide the one that never names you. Count each assistant separately, then look at the total.

Each assistant finds its sources its own way. Perplexity runs its own search, Gemini draws on Google Search, and ChatGPT's search is fed by OpenAI's own crawler, as our page on the assistants sets out. A brand that does well in Google can do well in Gemini and poorly elsewhere. Pooled together, a strong assistant can lift the total enough to mask a weak one.

The pooled figure also depends on how many answers each assistant gave. If one assistant failed on a few questions, the total leans towards the others. A per-assistant row, with its own counts, removes both problems.

What AI share of voice cannot tell you

Share of voice is only as good as the list of brands it counts and the questions it asks. Four limits follow from that.

• It only counts the competitors you track. Peec AI's page says it compares you against your tracked competitors, so a rival you never added is invisible to the figure.
• Adding a rival changes the figure. Put a new brand on the list and your share of voice falls, even though no answer changed.
• Questions that name you inflate it. An answer to "Is Example Co any good?" names Example Co almost every time, which says nothing about buyers who have not heard of you.
• It moves on its own. Assistants answer differently each time, so a small change between two readings may be noise.

The fourth limit is well measured. SparkToro's study, published in January 2026, had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI a combined 2,961 times. It found less than a 1 in 100 chance that ChatGPT or Google's AI would return the same list of brands in any two responses. The study concluded that a visibility percentage over many runs is a reasonable measure, and that a ranking position in AI is not.

The first limit is the one most often missed. If your tracked list was written from memory, it may leave out the brand the assistants actually recommend most. Your share of voice would then look healthy against rivals who are not your real competition in these answers.

None of these limits makes share of voice useless. They mean it answers a narrow question: among the brands on this list, on these questions, how much of the talk is about us? Read it as that, and it is a fair measure of standing against named rivals.

How to measure AI share of voice, step by step

You can measure share of voice by hand with a spreadsheet, or with a tool. The steps are the same either way.

1. Write the questions a buyer asks before they know your name, such as "best oven for a small bakery". Keep questions that name you on a separate list.
2. Pick the assistants to ask. ChatGPT, Perplexity and Gemini each search the web differently, so count each one apart.
3. Ask every question more than once, because one answer is a sample of one.
4. For each answer, write down every brand it names, not only the ones you expected.
5. Build the competitor list from those answers. Any brand named often belongs on it, whether or not you think of it as a rival.
6. Count your mentions and every tracked brand's mentions. Divide yours by the total.
7. Write the result with both counts, as in "12 of 80 mentions", and put your visibility count beside it.

Steps 4 and 5 are the ones that keep the figure honest. They make the answers decide who your competitors are, instead of your own assumptions. Step 3 matters almost as much: with each question asked three times, a brand that appears once is a weaker signal than one that appears every time.

Once you have the numbers, presenting them is its own job. Our guide on how to report AI visibility honestly covers the rules for that, such as keeping branded questions apart and calling a change only when it beats the normal swing. This post does not repeat them.

Where Agentcites fits

Agentcites does steps 2 to 6 for you. Today it asks ChatGPT, Perplexity, Gemini and DeepSeek up to 50 buyer questions each week and keeps every answer word for word. It shows share of voice beside the plain count of non-branded answers that named you, and keeps branded questions separate. Its competitor discovery learns your rivals from who keeps being named in the answers, rather than from a list you typed.