How to measure your brand’s share of voice in AI answers
Share of voice in AI answers is simple to define: of all the answers to the questions your customers ask, the share that offer your brand as an option. Measuring it properly comes down to what you count and what you leave out.
What to count
- Answers, not questions. Ask each question several times on each assistant. Every valid answer is one unit.
- Offered as an option, not mentioned in passing. “Unlike X, which is expensive…” is a mention of X, not a recommendation. Count a brand when the answer puts it forward as a choice.
- Only questions that do not name you. “Is X any good?” will always mention X. Those questions are useful for checking facts, and they do not belong in the share.
- Your competitors, from the same answers. The number means something only next to theirs.
A worked example
From a first scan we ran for an established professional-services firm: 19 customer questions, four assistants, 76 valid answers.
| Brand | Answers offering it | Share of voice |
|---|---|---|
| Competitor A | 29 of 76 | 38% |
| Competitor B | 25 of 76 | 33% |
| Competitor C | 16 of 76 | 21% |
| The firm | 7 of 76 | 9% |
The shares add up to more than 100% because an answer usually offers several brands. That is normal. Share of voice here is not a pie.
Break it down before you act on it
The single number hides most of what matters. The same firm was named in 21% of answers on one assistant and in none on another, in 25% of comparison questions and in none of the questions that start from a problem. Each of those points to a different piece of work.
Five mistakes that make the number meaningless
- Reporting a percentage without the count. “50%” of two answers and of two hundred are different facts. Always give both.
- Changing the questions between measurements. Then you are comparing two different things. Keep the set fixed for the period.
- Mixing assistants with different behaviour into one number without looking at each. One that searches the web and one that answers from memory move for different reasons.
- Counting your own name questions. It inflates the share and tells you nothing.
- Reading one day as a trend. Answers vary from run to run. A share becomes a trend only when the same questions are asked on a consistent schedule over weeks.
What a good number looks like
There is no universal benchmark: it depends on how crowded your category is. The useful comparison is with the leader in your own answers, and with yourself last month. NUR reports both, with the counts, by assistant and by type of question.