How to measure share of voice in AI search
Share of voice in AI search is your brand mentions as a proportion of all brand mentions across the same responses — the most stable comparative metric available.
Share of voice is the most useful comparative metric in AI search measurement, and the easiest to compute incorrectly. This article explains what it means in this context, how to calculate it properly, and the traps that produce misleading numbers.
In AI search, share of voice is your brand's mentions expressed as a proportion of all tracked brand mentions across the same set of responses. If a hundred responses contain three hundred brand mentions in total and forty-five of them are yours, your share of voice is 15%.
The reason to prefer it over a raw mention rate is stability. Engines change. A model update that makes answers shorter reduces every brand's mention rate at once, which looks like a decline for everyone and tells you nothing. Share of voice normalises that away: if all brands drop proportionally, your share is unchanged, correctly indicating that your relative position did not move.
How to calculate it
- Fix the prompt set. Every brand must be measured on identical prompts, run at the same time, on the same engines.
- Define the competitor set explicitly. Share of voice is only meaningful relative to a named list of brands, and changing that list changes the number.
- Count mentions per response per brand. Decide in advance whether repeated mentions of the same brand within one response count once or several times — counting once is more robust.
- Sum across all responses in the period.
- Divide your total by the total for all tracked brands.
- Report per engine as well as blended, since engines differ enough that the blend hides the story.
The traps
An incomplete competitor set. If you track three competitors and the engines routinely name eight brands, your share of voice is inflated by everyone you left out. Either include the full realistic set or label the metric clearly as share against a named subset.
Asymmetric prompt sets. Measuring yourself on your prompts and a competitor on theirs produces a number with no meaning. This sounds obvious and is a common defect in tools that bolt competitor tracking on afterwards.
Alias blindness. If you track your competitor's full legal name but engines use the short product name, you will systematically undercount them and overstate yourself.
Mixing text mentions with citations. Being named in prose and having your domain cited are different events. Combining them into one count produces a metric nobody can interpret.
Unstable prompt sets. Adding prompts mid-period breaks comparability with your own history. If you must change the set, restate the prior period on the new set or start a new series.
What to do with the number
Share of voice is a positioning readout, so use it for positioning decisions.
- Track it per question cluster. Your share on definitional questions and your share on comparison questions are usually very different, and only one of them relates to revenue.
- Watch the direction of the leader's share as well as yours. A leader losing share to a fragmenting field is a different situation from a leader consolidating.
- Use it to size the opportunity. If the top brand holds 40% and you hold 8%, the realistic near-term target is displacing whoever sits just above you, not the leader.
- Pair it with sentiment. Rising share with deteriorating sentiment usually means you are being named more often as the cautionary example.
Common questions about share of voice in AI search
- Why prefer share of voice over mention rate? Because it survives engine-wide changes. When a model update makes answers terser, every brand's mention rate falls and the chart implies a decline that did not happen. Share of voice correctly shows no change in relative position.
- Which tools measure share of voice in AI search? Any platform that tracks a defined competitor set on the identical prompt set at the same time. The requirement is methodological rather than a feature — competitor tracking bolted on afterwards frequently fails it.
- How many competitors should I include? Enough to cover the brands engines actually name, which is usually more than the three or four you would list from memory. Track your answers for a fortnight first and let the data define the set.
- Should repeated mentions in one answer count multiple times? Counting once per response is more robust and less sensitive to answer length. Whichever you choose, apply it consistently and document it.
- Can I compare my share of voice to a published benchmark? No. The number depends entirely on the prompt set and competitor set behind it, so cross-company comparisons are meaningless.
- What if a competitor is not tracked but keeps appearing? Add them, and restate the prior period if you can. An incomplete competitor set inflates your share and is the most common defect in these reports.
Tip: report share of voice with the competitor set and prompt count printed on the same slide. It is the fastest way to stop the recurring meeting argument about whether the number is real.
Still stuck? We typically reply within 1 business day.
Contact support