Share of Voice in AI Answers, Explained

Share of voice = your brand's mentions divided by total brand mentions across the sample, times 100. The definitions inside that formula matter. The sample must be a fixed panel of prompts, run the same way each period, or the number is not comparable over time. Mentions should count each answer once per brand, so one gushing paragraph does not outweigh presence across many answers. And weighted variants exist: counting a first-position mention more than a fifth-position one turns share of voice into a rank-sensitive score.

Keep the unweighted version as your headline number and the weighted one as a diagnostic. Mixing them silently is how teams fool themselves.

A raw mention rate answers "how often do we appear?" Share of voice answers "how much of the conversation do we own?", and the second question is the one that predicts pipeline. Suppose your mention rate rises from 20 to 30 percent of answers. Sounds like progress, until you learn the category leader appears in 95 percent and grew too, so your share of the total conversation barely moved.

Share of voice also normalizes across categories and prompt sets of different sizes, which makes it the right metric for comparing your position across providers, or explaining progress to an executive who thinks in market-share terms.

There is no universal benchmark, because share of voice depends on how fragmented your category is. Useful reference points instead: matching your real-world market share means AI reflects your position; trailing it badly means AI answers lag your actual standing, a gap worth closing fast. In a category with one dominant player, second place in share of voice with rising trend is a strong result. What you never want is a sustained zero, which usually signals an entity or crawlability problem rather than a competition problem.

Trend beats level. A share moving from 5 to 12 percent over two quarters tells you the strategy works; a static 20 percent tells you nothing is changing.

Three rules keep the metric honest. Fix the prompt panel and never quietly edit it, because adding prompts you win inflates the trend. Sample across providers separately, since your share in Perplexity and in ChatGPT are different numbers moved by different work. And log every competitor mention, not just your own, because the denominator is where the insight hides. The practical mechanics for a single provider are in how to see your share of voice in ChatGPT, and a free scan computes competitive share across five providers as part of its GEO score.