What is AI share of voice, and how is it measured?

AI share of voice is the percentage of all brand mentions in a set of AI answers that belong to your brand. You put the questions your buyers ask to assistants such as ChatGPT, Gemini and Claude, count every brand each answer names, and divide your brand's mentions by the total.

The idea comes from advertising, where share of voice usually means a brand's share of a category's media spend or impressions. In AI search, the "voice" is the answer itself. An assistant names a handful of brands instead of showing ten links, so the share of those names you hold is a direct measure of how often buyers hear about you before they visit any website.

In short

• AI share of voice = your brand mentions divided by all brand mentions, across a fixed set of questions, engines and runs.
• It is not the same as presence, the share of answers that name you at all. Report both.
• Calculate it on unbranded questions only. A question that names your brand inflates your share.
• Below about 30 answers, one answer moves the percentage too much to mean anything.
• Report each engine separately. One average across ChatGPT, Gemini and Claude can hide an engine that never names you.

How is AI share of voice calculated?

The arithmetic is simple. The discipline is in what you count.

1. Fix a question set. Use the questions buyers ask before they have a shortlist, such as "Which bank is best for a small business in South Africa?". Keep the set the same from one period to the next.
2. Collect answers. Put every question to every engine you track, more than once if you can. Each response is one answer.
3. List the brands in each answer. Count a brand once per answer, however many times that answer repeats it. Match name variants ("Meridian" and "Meridian Bank") to one brand.
4. Add up all the brand mentions. This total is your denominator.
5. Divide. Your brand's mentions divided by the total, times 100, is your AI share of voice.

Because you count each brand once per answer, your mentions equal the number of answers that name you. That makes the link between presence and share of voice easy to see.

A worked example (fictional brands)

For example, suppose Meridian Bank, a fictional bank, wants its AI share of voice for small-business banking. It runs 20 unbranded questions on three engines, which gives 60 answers. Every figure below is invented to show the arithmetic.

Brand	Answers naming it (of 60)	Presence	First position	Share of voice
Meridian Bank	24	40%	9 answers (15%)	24 of 150 = 16%
Competitor A	42	70%	30 answers (50%)	42 of 150 = 28%
Competitor B	30	50%	15 answers (25%)	30 of 150 = 20%
All other brands	54 mentions	n/a	6 answers	54 of 150 = 36%
Total	150 mentions		60 answers	100%

Three things stand out. Meridian is named in 40% of answers but holds 16% of the conversation, because the average answer names 2.5 brands (150 mentions across 60 answers). It is named first in only 9 answers, so when it does appear, it is usually not the lead recommendation. And Competitor A leads on every measure, which tells Meridian which rival to study.

Now split the same 24 answers by engine. Suppose Meridian appears in 12 of 20 answers on the first engine, 8 on the second and 4 on the third. The blended 40% hides the fact that one engine names Meridian in only one answer in five. That is where the work starts.

Which metrics belong next to share of voice?

Share of voice answers "how much of the conversation is ours?". It does not answer everything, so track it with:

• Presence: the share of answers that name you at all.
• First position: the share of answers that name you first, which is usually the recommendation a reader acts on.
• Cited sources: the pages each answer links to. These show where the answer came from, and which sites you need to appear on.
• Description: what the answer says about you. A mention that calls you expensive or out of date counts the same in the arithmetic as one that recommends you.

CitedAI shows, engine by engine, where you are recommended, which competitors are named instead and which sources shape the answers.

Sources are spread thin. In our analysis of 96 Google South Africa searches on 11 October 2026, the 86 AI Overviews that listed sources cited 10.9 sources each on average, drawn from 412 different domains. No single site owns a category's answers, which is why tracking the sources matters as much as tracking the mentions.

What makes an AI share of voice figure misleading?

• Small samples. With 30 answers, one answer moves a percentage by 3.3 points. With 10, it moves it by 10. If your "up five points" rests on 20 answers, it is one answer.
• Branded questions. "Is Meridian Bank good for small businesses?" names Meridian, so the answer will too. Mix such questions into the set and your share rises without a single new buyer hearing of you. Read more on branded vs unbranded questions.
• A changing question set. If you add or drop questions between months, the new figure is a new measurement, not a trend.
• One run per question. Ask an assistant the same question twice and it can name different brands. Repeat runs, or at least do not compare single runs.
• Blended engines. An average across engines hides the engine where you are missing.
• Counting repeats. An answer that names a brand five times is still one answer. Counting every repetition rewards long, repetitive answers.
• Look-alike names. A similarly named company is not you. Check that the mentions you count are your brand, not a namesake.

How should you report AI share of voice?

Keep the report boring and comparable:

1. One line per engine, never a blended average alone.
2. Unbranded questions only, with branded questions in a separate accuracy section.
3. Your share next to two or three named competitors, on the same questions.
4. The number of answers beside every percentage, so readers can judge the sample.
5. The same question set every month, with any change to it noted.

For the full method, see how to measure GEO without fooling yourself. For why share of voice matters in the first place, start with our guide to generative engine optimisation for South African brands.