Share of Voice

Share of Voice (SoV) is the competitive headline. It answers: when AI engines name brands like ours, how often is one of them us?

Definition

SoV(entity) = answers mentioning the entity / answers mentioning ANY tracked entity

Both numerator and denominator are counted at answer level: an answer that praises a brand five times counts once.

Stored per day × engine × entity. When no answer in the period mentions any tracked entity, SoV is undefined rather than zero — there was no voice to share, and a dash says so.

Tracked entities include your products as well as brands, so the same formula gives you product-level share when the Overview is in product scope.

Why answer-level presence

Because mention counting rewards verbosity rather than preference. Consider two answers:

• Answer A mentions you once, as the recommendation.
• Answer B mentions a competitor four times while explaining why it is unsuitable.

Counting mentions gives the competitor a 4-to-1 lead. Counting presence gives each brand one appearance, which is what actually happened: both were named once each, in one answer each. Tone is measured separately, by sentiment and recommendation strength.

Why it survives prompt-set changes

This is the property that makes SoV worth trusting.

Add fifteen prompts and your raw mention count rises — not because anything improved, but because you asked more questions. Visibility shifts too, because the denominator now includes the new answers.

SoV is more stable: both terms come from the same answers. If your new prompts perform like your existing ones, SoV barely moves. If they perform worse, SoV falls and tells you something true — you entered conversations you are not winning.

This is why SoV is the metric to put in a board deck, and why it is worth watching when you deliberately expand coverage.

Reading it

Against your own trend. Weekly movement on a stable prompt set is real signal.

By engine. SoV on Perplexity and SoV on Google AI Overviews are different competitions with different source economics. Averaging them hides where you are losing.

By prompt tag. SoV on comparison prompts is close to a leading indicator of competitive win rate. SoV on how-to prompts is a top-of-funnel measure. They should be read — and targeted — separately.

Alongside visibility. The two together diagnose the situation:

The denominator is a choice

Your competitor set defines the denominator. That has consequences worth stating plainly:

• Adding a competitor lowers everyone's SoV, including yours, with nothing having changed in the world.
• Removing one raises it.
• Tracking twenty brands makes every share small and every movement look like noise.

Three to six well-chosen competitors keep the metric legible. Record the date whenever you change the set, and be careful comparing across that boundary. See Competitors.

Read it with its interval

Like every rate in the product, SoV carries a 95% confidence interval, and changes are tested against a noise floor before being marked significant. A three-point shift on a small answer set is usually inside the floor.

Where the answer set is thin, the fix is more prompts on that subject rather than a bolder reading of the same evidence. See Metrics.

Weighted Share of Voice

The headline figure treats every mention as equal. A weighted variant refines it using two signals already collected:

• Position — the first brand named in an answer carries more weight than the fifth.
• Recommendation strength — being recommended outweighs being listed, which outweighs being warned against.

Weighted SoV rewards being the answer rather than an answer. Read the plain figure for coverage, the weighted one for competitive strength, and expect them to disagree when you are widely listed but rarely chosen.

Recommendation strength comes from the enrichment pass, which is not yet running — so the weighted variant arrives with it. See Product status.