Citino teamShare of voice in AI answers is your brand’s mentions divided by all brand mentions in the answers to the questions you track. It tells you how much of the conversation is yours when buyers ask ChatGPT, Google AI Overviews or Perplexity what to buy.
It is one of the four core numbers in AI search analytics, alongside visibility, position and sentiment.
Share of voice and visibility measure different things
| Visibility | Share of voice | |
|---|---|---|
| The question it answers | Were you named at all? | How much of the conversation is yours? |
| Counted as | Answers that name you ÷ all answers | Your mentions ÷ all brand mentions |
| Falls when | Answers stop naming you | Competitors are named more, even if you are not named less |
The two can move in opposite directions. If an engine starts naming five brands per answer instead of three, your visibility may hold while your share of voice drops, because the same mentions are now a smaller part of a bigger list. Tracking both shows which is happening.
How to calculate share of voice
- Fix the questions. Use the unbranded questions buyers ask, with the wording kept the same over time. How to choose the prompts to track covers this.
- Fix the competitor set. List the brands that count: you and the competitors that appear in answers to your questions. Brands outside the list are left out of both sides of the sum.
- Ask each engine on a schedule. The same questions, daily if you can, from a clean session.
- Count mentions per answer. Record every listed brand each answer names. Count a brand once per answer, however many times the answer repeats it.
- Divide. Your mentions ÷ the mentions of every listed brand, over all answers in the period.
Worked example (illustrative). Over 30 days you track 50 questions on Perplexity: 1,500 answers. Across them, the listed brands are named 4,000 times in total. Your brand is named in 600 answers. Your share of voice on Perplexity is 600 ÷ 4,000 = 15%, and your visibility is 600 ÷ 1,500 = 40%. The brands and numbers are fictional.
Choices that change the number
Share of voice is easy to inflate or deflate by accident. Decide these once, write them down and keep them fixed:
| Choice | Effect if done wrong |
|---|---|
| Including branded questions | Inflates your share, because those answers almost always name you |
| Changing the competitor list | Moves everyone’s share without any change in the answers |
| Counting repeat mentions within an answer | Rewards long answers that repeat a name, not being recommended |
| Weighting by position | Mixes share of voice with position; keep position as its own number |
| Pooling engines | Hides that you lead on one engine and are absent on another |
Measure it per engine
ChatGPT, Google AI Overviews and Perplexity read different sources, so your share of voice usually differs between them. Google says AI Overviews and AI Mode may issue multiple related searches across subtopics to build one response. Perplexity says it searches the web in real time and cites its sources. ChatGPT answers some questions from training data and searches the web for others, and its search answers come with citations to the pages used.
Report share of voice for each engine separately, and look at the cited pages behind each one. A low share on one engine usually points to a few sources that engine relies on and that do not mention you. How to find and fix source gaps covers what to do about them.
Put a margin of error on it
AI answers vary from run to run, so a share of voice from one week is an estimate. Report it with a margin of error, and calculate that margin over questions rather than over answers: answers to the same question tend to agree day after day, so treating each answer as independent makes the result look far more precise than it is.
The practical method is a bootstrap over questions: draw your questions at random with replacement, recompute share of voice, repeat many times, and take the middle 95% of the results, as the methodology describes. How to measure AI visibility works through the same method for visibility.
A share of voice without a margin of error cannot tell a real gain from a good week.
Reading a change in share of voice
- Smaller than its margin: not a change. Label it within noise.
- Everyone moved: if a model update or a change in how an engine searches shifted every brand at once, your change is not your doing. Compare it with the median competitor over the same days.
- One engine moved: look at what that engine cited before and after. The cause is usually in its sources.
To tell whether a specific fix raised your share of voice, measure the questions it targeted, net of the field’s change, against the margin of error. How to tell whether an AI search fix worked sets out the steps.
Where Citino helps
Citino is an AI search analytics tool. It asks ChatGPT, Google AI Overviews and Perplexity your buyers’ questions every day, with Gemini, Google AI Mode and Claude as add-ons, from a clean session with fixed wording, and counts every brand each answer names. Share of voice is reported per engine with the number of answers behind it and a 95% margin of error bootstrapped over prompts, as set out in the methodology. See how Citino measures.