What is Share of voice (AI)?
In AI visibility, share of voice is how often you are named in AI answers for your category's questions relative to your competitors — the answer-engine equivalent of market share of attention.
How is AI share of voice measured?
Fix a panel of the questions your customers actually ask, run each against the engines several times (answers vary between runs), and count appearances for you and each competitor. The output reads like polling: named in 59 of 60 runs; competitor A in 57; competitor B in 28. Repetition is what turns a lucky answer into a measurement.
Why measure share of voice instead of an 'AI rank'?
Because answers are non-deterministic — near-zero odds of identical brand lists across repeated runs — so a single-run 'rank' measures the run, not the brand. Rates across samples are stable enough to act on and honest enough to publish. When an answer happens to be a ranked list, the position within it is worth recording as an observed fact; it just isn't the headline number.
How do you actually grow share of voice?
Three compounding levers, none fast. On your site: extractability, so the passages engines lift can be yours. Off it: presence on the comparison pages and directories answers are assembled from. Underneath both: entity consistency, so systems are confident enough about what you are to name you. Expect movement over weeks, not days — which is why the measurement has to be a stable panel run on a schedule.
Frequently asked questions
What does Share of voice (AI) mean?
In AI visibility, share of voice is how often you are named in AI answers for your category's questions relative to your competitors — the answer-engine equivalent of market share of attention.
What is a good AI share of voice?
Benchmarks are young, but the working thresholds: appearing in a majority of category answers puts you in the conversation; parity with your strongest competitor means the answers treat you as a peer; dominance is being the default first name. Direction over weeks matters more than any absolute number.
How many prompts and runs make the measurement honest?
Enough repetition to separate signal from answer-to-answer noise — in practice panels of 20+ questions with 3+ runs each. One run of one prompt is an anecdote; the trend of a repeated panel is a metric.
Can share of voice be high while citations are zero?
Yes, and it's common: mentions flow from training data while citations require your pages to be retrieved live. A brand can be named in nearly every answer whose sources never include its own site — visibility built entirely on third-party pages.