AI Search Share of Voice: Measure Brand Mentions in Generative AI Answers for Visibility
How to calculate share of voice in AI answers: fixed prompt set, per-engine split, mention vs citation, position, sentiment, cadence. Then the tools that can run it, ranked.
Share of voice in AI search is the fraction of brand mentions you hold inside generated answers to a fixed set of questions, compared with a fixed set of competitors. That's the whole definition. Most of the arguments about it come from people changing one of those two fixed sets without saying so, and then comparing this month's number with last month's.
This post sets out the method first. After that it ranks the tools on our directory by how well they can execute it, because the method and the tooling fail in different places.
The method, step by step
Fix the prompt set. Pick the questions a buyer in your category would actually ask, grouped by topic. Thirty to a hundred is a workable range for one product line. This set is your denominator. If you add twenty prompts mid-quarter, your share moves even though no engine changed its answers. Freeze the list per reporting period and version it.
Fix the competitor set. Name the brands you're measuring against. Mentions of brands outside the list either get ignored or go into an "other" bucket. Decide which before the first run, because it changes the math.
Count per engine. Run every prompt on every engine you care about and keep the results apart. ChatGPT and Perplexity retrieve differently and cite differently. An average across engines hides the one where you're losing.
Do the arithmetic. Mention share is your mentions divided by all mentions of tracked brands across the answers for that engine. A made-up round example: forty answers name your four tracked brands a hundred times in total, and you account for twenty-five of them. Your mention share on that engine is 25%. A second, simpler number is presence rate: the share of answers that name you at all. Report both. Presence rate moves first when you start appearing. Mention share moves when you start displacing someone.
Split mention from citation. A mention is your brand named in the answer text. A citation is your URL in the sources. They behave differently. You can be named on the strength of a review site's page while your own pages never get cited. Track citation share as its own column, and record which URLs got cited, yours and third parties'.
Record position. Being named first in a list of five is not the same as being named fifth. Keep a first-mention rate next to share rather than inventing a weighting formula nobody can audit.
Attach sentiment. A mention that frames you as the expensive option still counts toward share. Report share with a sentiment split so a rising number with a souring tone doesn't read as a win.
Sample on a cadence. The same prompt can return a different answer an hour later. One snapshot is noise. Daily checks, read as a trend over weeks, are the minimum for a number you put in front of a CMO.
Segment where it matters. If you sell in several countries, or to buyers with very different needs, run the set per country and per persona. A share that blends Amsterdam and Austin hides two different competitive fields.
How we ranked the tools that can run it
The test for each tool: can it hold a frozen prompt set and competitor list, report share per engine, keep mentions and citations apart, sample often enough to trend, and segment by market or persona? Price matters only after that.
The ranking
1. Promptwatch. Verdict: every step of the method maps to a shipped feature. 2. Scrunch AI. Verdict: strong engine coverage, slow cadence. 3. Evertune. Verdict: the most rigorous sampling, priced for enterprise only. 4. Profound. Verdict: share of voice and citation share exist, but the wide engine list is Enterprise. 5. Brand24. Verdict: AI share of voice bolted onto social listening. 6. Peec AI. Verdict: per-prompt competitor share on a small model allowance.
Where each tool drops points
Promptwatch lists share of voice and competitive benchmarking on the same prompt data as sentiment analysis. Prompts carry search volume and difficulty scores, which helps when you're choosing the frozen set. Query fan-outs show the sub-questions an engine searches behind a prompt. Prompt trends show how visibility moved between checks and what changed. Citation analytics keep the mention and citation columns apart at page and domain level, with Reddit and YouTube citations broken out. Personas and country, state, or city targeting handle segmentation. Paid plans read the real product UIs across ChatGPT, Perplexity, Gemini, Claude, Google AI Overviews, and more. Where it drops points: Explore, the free tier, is ChatGPT only with 10 prompts, which is too small for a real competitor set. Essential at $95/mo gives you 50 prompts. Professional at $245/mo gives you 150.
Scrunch AI documents competitor benchmarking across up to nine engines, and Starter ($250/mo annual, $300 monthly) includes 350 custom prompts. The listing says data refreshes weekly, which is thin for a sampled metric. Prompt credits are consumed per engine, and G2 reviewers on the listing say reporting is weak enough that they export to Excel.
Evertune gets the sampling step right. The listing describes each prompt sampled around a hundred times and a share-of-recommendation measure. Pricing is custom (reported around $800/mo and up), onboarding takes weeks, and independent reviews are sparse.
Profound reports citation share, share of voice, and competitor cite-rates by prompt cluster. Starter at $99/mo annual is ChatGPT only with 50 prompts, so per-engine share isn't possible there. Growth at $399/mo annual covers three engines. The wider engine list sits on Enterprise, with contracts starting around $40k a year. G2 reviewers on the listing report citation counts that don't match manual ChatGPT checks.
Brand24 lists AI share of voice, sentiment, and key citing sources across nine engines. It's an add-on to Brand24 plans, and neither the add-on price nor the base plan price is published on the AI visibility page.
Peec AI tracks competitor share per prompt. Each tier includes three models, with extra models at $35 to $165/mo each, and there's no historical backfill, so your baseline starts on the day you subscribe.
Running the method in Promptwatch
Here's how the steps above translate, in order. Build the frozen set in prompt tracking, using volume and difficulty to drop prompts nobody asks. Tag prompts by topic so you can report share per product line. Add competitors to competitive benchmarking and read share of voice per engine, not blended. Open citation analytics to see whether your share comes from your own pages or from a third party's. Use prompt trends to read the number weekly and to see what changed when it moves. Turn on personas or city-level tracking only where your market actually splits.
The reason we point people to Promptwatch for this is the citation column. Most trackers will hand you a share number. Far fewer show which URLs earned it, and that's what tells you what to fix. You can start on promptwatch.com with Essential and expand the prompt set once the first month's trend is readable.
Mistakes that wreck the number
Changing the prompt set without versioning it. Averaging engines. Counting your own brand's mentions in prompts that name you ("is [brand] good"), which inflates share with answers you asked for. Reporting share without a sentiment split. Comparing a weekly-refresh tool's number with a daily-refresh tool's number as if they were the same measurement. Each of these produces a cleaner chart and a less accurate one.