How to measure AI share of voice: a step-by-step method
How to measure AI share of voice: mention, recommendation and citation share across ChatGPT, Gemini, Perplexity and Copilot, with worked maths.
What “share of voice” means in AI search
In advertising, share of voice is your share of all the advertising in a category. In SEO, it usually means your share of estimated clicks across a keyword set. In AI search there are no positions or click estimates to sum — so AI share of voice is measured by counting appearances in sampled answers.
It is most useful when split three ways:
| Metric | Counts | Answers the question |
|---|---|---|
| Mention share | Answers that name the brand | “Are we part of the conversation?” |
| Recommendation share | Answers that suggest the brand as an option | “Are we on the shortlist?” |
| Citation share | Cited sources that are the brand’s own pages | “Is our content being used?” |
Step 1: Choose the competitive set
Pick the four to eight brands buyers actually compare you with. Include the category leader even if you don’t see them as a direct competitor — buyers do.
Step 2: Build the prompt set
Write 20–60 prompts across category (“best X for Y”), comparison (“X vs Y”), problem (“how do I…”) and brand questions. Phrase them the way buyers speak. Keep the set stable; add new prompts as a separate cohort so trends stay comparable.
Step 3: Sample consistently
Run each prompt at least three times per platform per measurement period. Keep conditions consistent: same account state (logged out where possible), same location, same mode (search on or off). Record the date and platform for every run.
Step 4: Code each answer
For each answer, record for every tracked brand: mentioned (yes/no), recommended (yes/no), position of first mention, and which domains were cited. A spreadsheet works; consistency of coding rules matters more than tooling.
Step 5: Do the maths
Suppose you ran 30 prompts × 3 runs × 4 platforms = 360 answers.
- Your brand was mentioned in 108 answers. Across all tracked brands there were 540 brand mentions in total. Mention share = 108 ÷ 540 = 20%.
- Your brand was recommended in 54 answers; all brands together were recommended 300 times. Recommendation share = 54 ÷ 300 = 18%.
- Answers cited 1,200 sources in total, of which 84 were your pages. Citation share = 84 ÷ 1,200 = 7%.
You can also report a simpler presence rate: your brand appeared in 108 of 360 answers = 30%.
These numbers are a worked example, not benchmarks.
Step 6: Read the results carefully
- Compare like with like. Only compare periods measured with the same prompts, platforms and run counts.
- Respect variation. With small samples, a move from 18% to 21% may be noise. Look for sustained trends across several periods.
- Segment. Share of voice by platform and by prompt type usually tells you more than the total.
- Pair with outcomes. Branded search trends, AI referral traffic and “how did you hear about us?” answers connect visibility to business results.
What AI share of voice can’t tell you
It doesn’t measure how many people saw an answer — platforms don’t publish that. It reflects your prompt choices, so a biased prompt set produces a biased result. And it won’t capture personalised answers for logged-in users. State these limits whenever you report the number.
Where first-party data helps
Some platforms now report part of the picture. Google includes AI Overviews and AI Mode clicks in Search Console’s Web totals, and Bing Webmaster Tools’ AI Performance report (public preview) shows how often your pages are cited in Copilot and Bing AI answers. Use these alongside sampling, not instead of it.
For the broader measurement framework, see how to measure AI visibility; to have it run for you, see our AI visibility service.