Why it matters
Ask an assistant the same question twice and you can get two different lists of brands. One answer, or one screenshot, tells you almost nothing; a rate over many answers tells you how often you appear, and an interval around it tells you how much to trust that number.
Without both, a team reacts to noise: a “drop” that is two answers out of twenty, or a “win” that disappears next week. Measuring this well is mostly a matter of asking enough times and being honest about the uncertainty.
The workflow
Fix the question set first
Measure the same tracked questions every time, per market. If the list changes between runs, the numbers are not comparable and every trend is suspect.
Ask each question many times, on each surface separately
Assistants do not give the same answer twice, so repeat each question and record, per answer: was your brand mentioned, was it recommended, was your domain cited, and which other brands were named. Keep the surfaces apart. OpenAI’s models, Gemini, Perplexity and Google’s AI Overviews retrieve different pages and behave differently.
Turn the tallies into rates with an interval
“Mentioned in 12 of 40 answers” is a 30% rate with a wide interval. Show a 95% credible interval next to every rate (a Beta distribution over hits and misses is the standard way to get one). With ten answers the interval is wide; that is information, not a flaw. Say “the probability an answer mentions you”, never an exact rank.
Compare against the brands you compete with
Share of voice is your mentions out of all tracked brands’ mentions. Also track whether you are listed first (only in answers that give an ordered list), the prompts where you never appear, and the answers where your domain was retrieved but not cited.
Decide in advance what counts as a change
Set the bar before you look: for example, a move of at least 10 points, at least 10 answers in each window, and at least 85% probability that the change is real. Anything smaller is reported as “no significant change”.
Track it weekly in VozRadar
VozRadar runs a baseline over your approved prompts, then scheduled weekly tracking, and shows every rate with its interval on the Overview page.
Run it in VozRadar
Overview
- Rates of answers that mention, recommend or cite your domain, each with a 95% credible interval shown as a whisker
- Share of voice against tracked competitors, “listed first” for ordered answers only, prompts where you never appear, and your domain retrieved but not cited
- Flags a significant change only at 10 points or more, with at least 10 answers in each window and at least 85% probability
- Results per answer surface and per market, from a baseline, weekly scheduled tracking or manual runs
In the app: Project → Overview. OpenAI, Gemini, Perplexity and Claude are measured through their official APIs; Google AI Overviews and AI Mode from Google’s own results pages. Claude and the Google surfaces start with the Growth plan.
See how AI sees your brand.
Start with one market. Expand when you're ready.