AI share of voice tools: how each one counts, and how not to be fooled

What AI share of voice tools measure

AI share of voice tools measure your slice of the brand mentions in AI answers. They ask assistants a set of questions, count every brand named, and report your mentions as a share of the total.

Think of a pie. Every time an answer names a brand, that brand gets a slice. Your share of voice is the size of your slice compared with the whole pie. The catch is that the pie only contains the brands the tool was told to look for, and each tool cuts it a little differently.

That is why two tools can show the same brand at very different shares, with both doing the arithmetic correctly. The rest of this guide shows where the differences come from, so you can read any share of voice figure and know what it is really saying.

Share of voice is a share of the brands you track, not a share of the market. Change the list and the share changes.

How the main tools define share of voice

The tools publish different formulas. The table quotes each vendor's own documentation, read on 10 October 2026, and marks where a definition was not stated on the page we read.

Peec AI's version is the simplest to recompute. Its documentation gives the formula as your brand's mentions divided by the total number of brand mentions, times 100 (Peec AI docs).

Semrush adds weights. Its July 2026 guide says the Brand Performance report bases share of voice on how many times your brand is mentioned and how high it appears, and that for ChatGPT the calculation also factors in the topic's search volume (Semrush blog). A weighted figure can be more useful for some purposes, but it is harder to check by hand, because you need the weights as well as the counts.

The questions differ too, and that changes the figure as much as the formula does. Ahrefs says Brand Radar takes People Also Ask questions from its keyword index and enters them into the web version of each chatbot using the default model (Ahrefs help). A tool built that way measures you across a broad pool of public questions. A tool that asks only the questions you wrote measures you on your own buyers' questions. Neither is wrong, but a share from one cannot be compared with a share from the other.

Share of voice and visibility are different numbers

Visibility asks how many answers named you. Share of voice asks how big your slice is among all the brands named. They can move in opposite directions, so a tool that shows only one of them can tell a misleading story.

Peec AI's documentation gives a clean example. If your brand is mentioned in 4 of 10 chats, your visibility is 40%. If a competitor is mentioned 12 times in those same chats, your share of voice is 25%, because 4 divided by 16 is a quarter (Peec AI docs).

The table below is a made-up example that shows how the two can split. Between week one and week two, the assistants start naming fewer brands per answer. Your share of voice rises, yet fewer answers mention you.

Neither number is wrong. They answer different questions. Visibility tells you how often a buyer who asks will hear your name. Share of voice tells you how crowded the answers are around you. A report that shows share of voice alone can celebrate a week in which buyers heard about you less.

Four ways a share of voice figure fools you

Most misleading share of voice figures come from what is in the denominator, not from bad arithmetic. These four are the ones to check first.

1. The tracked list sets the pie

Share of voice divides by the mentions of brands you track. Peec AI's documentation describes it as how often your brand is mentioned compared to all of your tracked competitors (Peec AI docs). Leave out a strong rival and your share goes up. Add one and it goes down, even if every answer stayed the same.

2. Questions with your name inflate it

Ask an assistant about your product by name and it will name you back almost every time. If those branded questions sit in the same pool as buyer questions that never mention you, they lift your share without one new buyer hearing about you. Report them apart.

3. Position weights add noise

A formula that rewards appearing early in an answer depends on the order of brands, and that order is close to random. A SparkToro and Gumshoe study of 2,961 prompt runs found less than a 1 in 1,000 chance that ChatGPT or Google's AI gave the same brands in the same order twice (SparkToro).

4. Small samples swing

The same study found the list of brands itself differed between almost every pair of responses, while how often a brand appeared across many runs was steadier (SparkToro). A share built on a few dozen answers can move several points in a week with nothing changed. Ask the tool how big the sample is behind each figure.

When share of voice is the right number to watch

Share of voice is the right number when you want to know how you stand against a known set of rivals on the same questions. It is the wrong number when you want to know whether buyers hear about you at all.

It works best in a settled category. If the same five brands keep appearing, a share of those five tells you whether you are gaining ground on them. A rise from one slice in ten to one slice in eight is meaningful when the pie is stable and the questions have not changed.

It works worst in a young or crowded category, where new names appear every month. Each newcomer that enters the tracked list shrinks everyone's share, so a falling figure can mean the category grew rather than that you lost ground. In that case, lead with visibility and treat share of voice as background.

• Use share of voice to compare yourself with named rivals on a fixed question set.
• Use visibility to answer whether a new buyer hears your name at all.
• Use both together when you report, each with its counts, so neither can tell its story alone.
• Restart the baseline whenever the tracked list or the question set changes.

There is one more reason to keep visibility in front. A buyer reads one answer, not the pie. What matters to that buyer is whether your name is in the answer they got, and visibility is the figure that measures exactly that.

What to ask before you buy an AI share of voice tool

Ask how the figure is built before you ask how it looks. A tool that answers these plainly is one you can defend to your own boss.

1. What exactly does the share divide by: mentions, answers, or a weighted score?
2. Which brands are in the pie, who chose them, and can I add a rival the tool has missed?
3. Are questions that contain my name kept out of the figure, or reported separately?
4. Does the figure weight position or search volume, and can I see it without the weights?
5. Can I open the answers behind any figure and read them word for word?
6. How many answers sit behind each figure, and how much does it move when nothing changes?
7. Which assistants are asked, how often, and through the public app or an API?

If you are comparing vendors, our comparison pages set out what each one says about itself, each fact quoted from the vendor's own site with the date it was read.

How to measure AI share of voice yourself

You can work out share of voice by hand from the same sheet you use for brand mentions. It takes longer than a tool, but you will know exactly what the figure means.

1. Fix a list of buyer questions that do not contain your name, and keep it the same each week.
2. Ask each question on each assistant you care about, and save every answer in full.
3. Write down every brand each answer names, not only the ones you expected.
4. Count your mentions and the total mentions of every brand on your list.
5. Divide the first by the second, and write both counts beside the percentage.
6. Work out visibility too: the answers that named you over the answers you asked.

Step three is the one tools often skip. If you only count brands you already know, a new rival that starts winning answers never enters the pie, and your share looks healthier than it is. Our guide to tracking brand mentions in AI search has a sheet layout you can copy.

Here is the arithmetic with made-up numbers. Say 25 questions on four assistants give 100 answers. Across them, your brand is named 18 times and the brands on your list are named 150 times in total, yours included. Your share of voice is 18 of 150, or 12%. Your visibility is the answers that named you, say 16 of 100, because two answers named you twice. Write both lines in the report, each with its counts.

Where Agentcites fits

Agentcites reports visibility first: the share of non-branded answers that named you, with both counts beside it, per assistant. It is not a blended score. Questions with your name are reported apart, and every figure opens the answers it came from, as the answer tracking page explains.

It also learns rivals from who keeps winning your questions, rather than only from a list you type in, which keeps the pie closer to the real field. It has a limit here too. A brand that wins only once in a scan is not added to the tracked list, so it shows only in the question you lost. Agentcites asks ChatGPT, Perplexity, Gemini and DeepSeek every week, for $149 a month per website.