Share of voice in AI search is the share of answers where your brand is one of the names given back. It is not traffic, and it is not rank. It is how often the assistant decides you belong in the set it hands a buyer.
Count the right thing
A mention is not a recommendation. An assistant naming you in a list of eleven options is not the same as it saying you are the one to call. Separate the two from the first day you measure, because only one of them puts a buyer in front of you, and a chart that blurs them will tell you things are fine while nothing sells.
Sample enough to trust it
Assistants answer the same question differently each time. Ask once and you have an anecdote. The baseline most teams need is a fixed set of buyer prompts, run on a schedule, across every assistant your buyers actually use, with enough samples per run that a move is bigger than the noise.
That gives you a confidence band rather than a number, which is the honest form of this metric.
Table stakes
- A named set of prompts a real buyer would type, not keywords.
- Every assistant your market uses, asked the same way, so the four can be compared.
- Mention and recommendation tracked apart.
- A fixed cadence, so week-on-week movement means something.
- At least one competitor set, because share of voice is relative by definition.
If you have those five, you can answer "are we getting recommended, and is it improving". Most teams reporting on AI visibility cannot.