Share of voice in AI answers: how it is calculated and how to read it

Visibility and share of voice are not the same

Visibility, sometimes called inclusion rate, answers a yes or no question for each answer: was the brand named? Averaged over many answers it becomes a percentage, such as 'named in 40% of answers'.

Share of voice answers a different question: of all the brand mentions made in these answers, how many were yours? It captures the effect of crowding. In a category where each answer names eight brands, being named in 40% of answers may still be a small slice of the total conversation. In a category where answers name two brands, the same 40% is dominant.

Use both. Visibility tells you how often you get through the door. Share of voice tells you how you stand once in the room.

The arithmetic

Take a group of prompts, run each several times across an engine, and collect the answers. Count every mention of every brand in those answers. Divide your mentions by the total.

Example: 50 answers. Across them, brands are mentioned 220 times in total. Your brand is mentioned 33 times. Your share of voice is 33 / 220 = 15%. Your visibility is separate: if you appear in 24 of the 50 answers, visibility is 48%.

Decide in advance what counts as a mention and keep the rule fixed. The common choices are: count each brand once per answer (a presence-based measure), or count every occurrence (a frequency-based measure). Presence-based is more stable, because a long answer that repeats a name six times does not outweigh a short one. Whichever you choose, say so in the report.

Related reading: How to measure AI visibility (a practical method) and Why confidence intervals belong in every AI visibility report.

Decide who is in the denominator

The result depends heavily on which names are counted. If the answers mention publishers, marketplaces, products and people as well as brands, including them would shrink everyone's share and blur the picture of who is competing for the buyer.

A clean approach is to count only entities that act as competing brands in the category: the companies, products or services a buyer could choose between. Exclude generic sources (a news site cited as a reference), places and people unless they are the thing being chosen.

Keep a tracked competitor list as well. The ranking of named competitors in your report is what makes share of voice actionable, because you can see exactly who is taking the share you do not have.

Why the number needs context

Sampling noise. Generated answers vary run to run. A share of voice of 15% from 12 answers could plausibly be 8% or 25%. Show the number of answers and a range, and be cautious about reading small changes.

Engine differences. Each engine has its own habits, such as how many brands it lists, whether it names brands at all for a question, and how readily it cites sources. Compare an engine with itself over time, not engines with each other, unless you adjust for these habits.

Prompt mix. Adding a group of prompts where you are strong will raise the overall number without any real change in the market. Keep a frozen baseline set for trend lines and add new prompts as a separate cohort.

Market. Answers differ by country and language. A brand's share of voice in India can be very different from its share in the United States, so report by market.

What to do when it moves

If share of voice rises, find out why before celebrating. Did a competitor drop out of the answers? Did a new source start citing you? Did the engine change how many brands it lists? Knowing the cause tells you whether the gain will last.

If it falls, check the answers themselves. Read a sample from before and after. Look at which competitors gained, which sources are now cited, and whether your pages are still being fetched. A drop is often traced to a change in what the engine cites, or to a crawler being blocked.

In both cases the sentences matter more than the percentage. The answer text shows the framing, the reasons the engine gave for recommending a brand, and the sources behind them. Those are the things you can act on.

Pairing it with the other measures

Share of voice should travel with three companions: average position when named (are you listed first or fifth), sentiment (are you recommended, mentioned in passing or caveated), and citation share (do the sources behind the answer belong to you or to someone else).

A brand with high share of voice, low citation share and weak sentiment is being talked about on borrowed evidence. A brand with moderate share of voice, strong citations and positive framing has a sturdier position and often grows more reliably.

OClever calculates share of voice on named brands only, keeps branded and unbranded prompts apart, and shows the number of answers behind every figure so a small sample is not mistaken for a trend.