AI share of voice is the proportion of brand mentions inside a set of AI answers that belong to you. Ten brands named across a month of answers, thirty of the three hundred mentions yours, and your share of voice is 10%.
It is the metric most often confused with visibility, and the confusion matters, because the two suggest opposite actions.
The formula
For a fixed set of prompts, over a fixed period:
share of voice = your brand mentions / all brand mentionsThree details decide whether the output is meaningful.
The prompt set has to be fixed. Add three prompts where you happen to be strong and your share of voice rises without anything changing in the world. This is the easiest way to produce a flattering number and the easiest way to fool yourself. Fix the set, then let it run.
The competitor set has to be defined. Whether an unrelated company that shares a word with your category counts as a mention changes the denominator. Whoever counts, counts consistently.
Repeat mentions need a rule. If an answer names you three times, is that one mention or three? Both are defensible. Counting once per answer is the more stable choice, because it stops a single verbose answer dominating a week.
Why it is not the same as visibility
Visibility asks: in what share of prompts was my brand mentioned at all? Share of voice asks: of the mentions that happened, how many were mine?
Consider two brands in the same category over the same forty prompts.
- Brand A appears in 8 of 40 answers, but in those eight it is the first named and often the only one. Low visibility, high share of voice.
- Brand B appears in 30 of 40 answers, always as the fourth item in a list of six. High visibility, low share of voice.
These are different problems. Brand A is trusted where it is known and simply is not known widely enough; the work is presence, in more of the category’s questions. Brand B is known everywhere and chosen nowhere; the work is differentiation, because being the fourth of six names is close to being invisible in a form that looks like success on a dashboard.
A single blended score hides which of these you are.
What share of voice does not tell you
Whether the mention helped. “Avoid X if you need compliance” counts as a mention. Share of voice is volume, not sentiment, and the two need reading together.
Where in the answer you appeared. First in a list and last in a list are worth very different amounts to a reader who stops after two names. Some teams track position separately for this reason.
Whether anyone can act on it. A mention with no link and no citation gives the reader nothing to click. It counts, and it converts nothing.
Reading it per engine
Averaging across engines is the most common way to lose the plot. Five engines will give you five different pictures of the same category, because they retrieve from different indexes and weight sources differently.
A blended 20% could be 20% everywhere, which is one situation, or 60% on one engine and near zero on four, which is a completely different one with a completely different fix. Always look at the per-engine breakdown before drawing a conclusion from the average.
How to use it
Share of voice is a comparison metric, so use it comparatively.
- Against yourself, over time. The trend on a fixed prompt set is the most honest number available to you.
- Against named competitors, in the same answers. Not against an industry benchmark, which was computed on someone else’s prompt set and is not comparable to yours.
- Segmented by prompt type. Share of voice on branded prompts (“what does Noma do”) is nearly meaningless; you will dominate. Share of voice on category prompts (“best tool for X”) is the number that reflects whether you are winning.
That last split is the one most worth building into your reporting. Branded and category prompts blended together produce a figure that flatters you and moves for reasons you cannot act on.