Share of voice in AI answers measures how often your brand shows up compared with competitors across a fixed set of prompts and engines. There are 2 common ways to calculate it. Mention rate asks how many answers name you. Share of mentions asks what portion of all brand names in those answers were yours. Pick one and keep it.
Two definitions, two numbers
Start with a set of prompts, the AI engines you track, and the competitors you compare against. Run every prompt on every engine, and record which brands each answer names.
Mention rate is the number of answers that mention your brand divided by the number of answers. If you appear in 24 of 60 answers, your mention rate is 40%. Each competitor gets their own rate, and the rates do not add up to 100%, because one answer can name several brands.
Share of mentions divides your mentions by the total mentions of all tracked brands. It always adds up to 100% across the brands you track, which makes it handy for a pie chart and a poor fit for judging whether you are visible at all. Share can stay flat while everyone’s mention rate rises or falls.
Neither is wrong. Mention rate tells you how visible you are. Share of mentions tells you how much of the conversation you hold against the others.
A worked example
This example uses invented numbers to show the arithmetic. It is not real data and does not describe any actual brand.
Suppose you track 20 prompts on 3 engines, which gives 60 answers. Four brands appear across them.
| Brand | Answers naming it | Mention rate | Share of mentions |
|---|---|---|---|
| Your brand | 24 | 40% | 26.7% |
| Competitor B | 36 | 60% | 40.0% |
| Competitor C | 18 | 30% | 20.0% |
| Competitor D | 12 | 20% | 13.3% |
| Total | 90 | not applicable | 100% |
Mention rate is the second column divided by 60. Share of mentions is the second column divided by 90, the total of all four. The 90 is larger than 60 because many answers named more than one brand.
Now split your 24 mentions by engine, still with invented numbers.
| Engine | Answers | Naming your brand | Mention rate |
|---|---|---|---|
| Engine 1 | 20 | 12 | 60% |
| Engine 2 | 20 | 8 | 40% |
| Engine 3 | 20 | 4 | 20% |
The blended 40% hides a 40-point spread. If you only watched the average, you would miss that one engine barely knows you and another favors you. This is why Doily shows share of voice per engine and per prompt, not only as a single score.
How noisy is it
AI answers vary from run to run. SparkToro and Gumshoe.ai had volunteers run 12 brand-recommendation prompts through ChatGPT, Claude, and Google’s AI features a combined 2,961 times. They reported that the chance of getting the same list of brands in any 2 responses to the same prompt was less than 1 in 100, and that the order of brands was even less stable.
The same research points to what holds up. Brands that the AI associates with a topic show up repeatedly, so how often you appear across many runs is a reasonable proxy for how visible you are. SparkToro’s author concluded that ranking position inside an answer is not worth tracking, while the percentage of responses that include a brand has some statistical validity. The authors also say more research is needed on how many runs are enough.
For intuition, here is a rough calculation using ordinary statistics, not anything from that study. A 40% mention rate measured on 60 answers has a margin of error of roughly 12 points either way, so the true rate could plausibly be anywhere from about 28% to 52%. That assumes the answers are independent, which they are not quite, so treat it as a lower bound on the uncertainty. The practical rule: on a sample this size, a difference of 5 points between you and a competitor, or between this week and last week, is not a finding.
Three things shrink the noise:
- Run each prompt more than once per period and count appearances across the runs.
- Use more prompts. 30 well-chosen prompts is a workable start for a single topic.
- Compare averages over weeks rather than single days.
Which one to report
For a business owner, lead with mention rate per engine. “When someone asks this question on this engine, we come up in 4 of 10 answers” is easy to picture and easy to compare month to month. It also does not move when a competitor changes, which makes it a cleaner read on your own work.
Use share of mentions when the conversation is about the competitive set: who is gaining ground, who is losing it, and by how much. A shift from 26.7% to 31% of mentions means something even when the mention rate barely moved, because it says the others lost ground.
If some prompts matter more than others, weight them. A prompt that sounds like a customer ready to book should count for more than a general question about the topic. Write the weights down before you look at results, so the weighting cannot drift toward whatever flatters you.
Rules that keep the number honest
- Fix the prompt set. If you add prompts that favor you, share of voice rises and nothing real changed. Add prompts deliberately and record the date.
- Keep the engines constant. Compare like with like. An engine added in March changes the March total.
- Name the competitors up front. Choose the 3 to 5 brands you actually lose business to, not whoever shows up most.
- Hold location steady. Answers can change by place, so a fixed city or country belongs in the definition.
- Separate mention from citation. A mention names you. A citation links to your site. Share of voice here counts mentions. Track citations as their own measure.
Doily tracks mentions, citations, and share of voice versus competitors for each engine and prompt, by country or city. Paid plans check prompts daily, and Free checks weekly, so the sample grows on its own schedule. Our guide to reading share of voice walks through the dashboard version of this arithmetic, and the guide on tracking AI prompts covers building the prompt set.
What it cannot tell you
Share of voice says how visible you are. It does not say whether the mention was favorable, whether it led to a visit, or whether it moved revenue. A brand can have a high share because people complain about it. Pair the number with the text of the answers, because the sentence an engine writes about you matters more than the count.
