FUNDAMENTALS · 6 MIN READ
What “share of voice” means when there's only one answer
In AI answers, share of voice is the percentage of sampled answers in which your brand appears, across a fixed prompt set, over repeated runs. There is no page two: for any single answer, your visibility is binary. The metric only becomes meaningful in aggregate - many prompts, many runs, tracked over time - because answer engines are stochastic and one answer tells you almost nothing.
Search built a generation of metrics on the assumption of a results page: ten positions, measurable click-through at each, partial credit for ranking anywhere. An answer engine breaks that assumption. The user asks, the engine writes one answer, and the brands inside that answer are the only brands that exist for that user in that moment. Everyone else got zero.
Binary per answer, continuous in aggregate
The way to recover a usable metric is sampling. Ask the same prompt many times, across the prompts that matter, and count: in what fraction of answers does your brand appear? That fraction is your share of voice. It behaves like a probability, because it is one - an estimate of the chance a real buyer asking a real question sees you at all. Run one prompt once and you have a coin flip's worth of information. Run a prompt set on a schedule and you have a trend line.
Appearance is not the whole story
Two brands can both appear in 60% of answers and have very different businesses. One is named first and recommended; the other appears in a but-consider-alternatives clause. So we split the metric into layers: presence (are you in the answer), position (first mention, or fifth), and framing (recommended, listed neutrally, or hedged against). A presence number alone flatters brands that appear as the cautionary example.
Why the stochasticity is a feature of the measurement
It's tempting to treat answer variance as noise to engineer away. It isn't noise - it's the actual distribution your buyers experience. Different users asking the same question get different answers, so the honest metric is a rate, not a screenshot. This also sets the bar for what counts as change: if your presence rate moves from 42% to 45%, you likely watched sampling error, not progress. ClerAEO's rule is that movement under five percentage points at the day-28 mark is reported as no change. That threshold costs us the ability to celebrate small wins, and it saves customers from celebrating fake ones.
Practical implications
- Fix the prompt set before you measure. A share-of-voice number is only comparable to itself over time if the denominator stays constant.
- Sample each prompt multiple times per period. Single runs produce trend lines made of dice rolls.
- Track engines separately. Your ChatGPT presence and your Claude presence are different numbers moved by different mechanisms - we track those two live today, with more engines rolling out.
- Report position and framing next to presence, or the number will mislead someone in your own company.
The one-answer world is less forgiving than the ten-blue-links world, but it's more honest about what marketing always was: a fight to be the answer, not a fight to be on the list. Measure it like a probability, judge it over weeks, and distrust any single answer - including a good one.
Frequently asked
How do you calculate share of voice in AI answers?
Fix a prompt set, run every prompt many times per period against each engine, and count the fraction of sampled answers in which your brand appears. That fraction is the metric. It behaves like a probability because it is one - an estimate of the chance a real buyer asking a real question sees you at all.
How much movement in an AI visibility number is real?
Less than you would like. A move from 42% to 45% is usually sampling error rather than progress. ClerAEO reports movement under five percentage points at day 28 as no change, which costs the ability to celebrate small wins and prevents celebrating fake ones.
See where you stand.
Run a free scan and get your own answer-engine scorecard.