AI share of voice measures how often a brand appears in a defined sample of AI-generated answers compared with competitors. It can help a marketing team understand whether ChatGPT, Perplexity and other assistants recommend the business for actual buyer questions. It is not a Google ranking, a universal market-share number, or a measure of revenue.
This guide focuses on the calculation, the sampling decisions that make results comparable and the practical actions a team can take. For the broader distinction between being mentioned and being cited, see our AI brand mentions vs citations explainer.
AI share of voice: definition and formula
Choose one clearly defined prompt set, one AI platform and one measurement period. Then calculate:
Brand mention rate (%) = (answers naming your brand ÷ valid answers tested) × 100.
When comparing brands, you can also calculate a competitive share of mentions: your brand mentions divided by the sum of all tracked brands’ mentions in the same answers, multiplied by 100. An answer can mention several brands. Therefore total mentions can exceed the number of answers; always label which denominator you use.
Worked example: three CRM vendors (illustrative only)
A hypothetical B2B team tests 20 buying-intent prompts. Its tool records 8 answers mentioning Brand A, 12 mentioning Brand B and 5 mentioning Brand C. Some answers mention more than one brand.
| Tracked brand | Answers containing brand | Mention rate (20 answers) | Share of 25 tracked mentions |
|---|---|---|---|
| Brand A | 8 | 40% | 32% |
| Brand B | 12 | 60% | 48% |
| Brand C | 5 | 25% | 20% |
All figures above are invented solely to explain the calculation; they are not client or market results. A single platform or 20 prompts cannot represent the entire category. Use consistent sampling and describe uncertainty before drawing conclusions.
How to design a prompt set that reflects commercial intent
- Choose 3–5 buyer journeys: discovering a category, comparing suppliers, checking alternatives, evaluating fit, and deciding where to enquire.
- Write 5–10 natural questions per journey: include company size, budget, geography and use case where appropriate.
- Separate branded prompts: asking for a brand by name tests different visibility than unprompted supplier discovery.
- Save the test conditions: platform, date, language, country, account/session conditions and exact prompt text.
- Repeat the same panel: evaluate changes directionally, and create a separate panel for emerging questions rather than silently changing the sample.
Measure more than a brand mention
| Metric | What it tells you | Common mistake |
|---|---|---|
| Unprompted mention rate | Frequency of brand appearing for non-branded questions | Mixing in prompts that already contain the brand |
| Cited-domain rate | Frequency your website appears as a linked source | Calling a citation a recommendation |
| Recommendation position | Where your brand appears within an answer, if meaningful | Treating the order as a stable search ranking |
| Factual accuracy | Whether claims about your business are correct | Counting incorrect mentions as success |
| AI referral sessions | Identifiable website visits from AI sources | Assuming all AI-influenced visits are trackable |
| Qualified enquiries | Commercial value of visits and recommendations | Replacing CRM outcomes with a visibility score |
For a practical approach to referral sessions and conversions, use our GA4 AI traffic guide. For the scope of a baseline exercise, use the GEO audit checklist.
A realistic 30-day GEO measurement plan
Days 1–7: define 30 commercially relevant prompts and three direct competitors; record baseline answers and sources. Days 8–14: identify the specific comparison pages, product facts and third-party sources represented in those answers. Days 15–21: improve the most relevant existing page and correct public company profiles. Days 22–30: retest the same prompts, capture screenshots or source exports, and report directionally. If you need statistical confidence, increase sample size and account for repeated responses.
What if mentions increase but leads do not?
Audit the prompt mix. You may be appearing for educational questions unrelated to purchases. Review landing-page conversion paths, answer accuracy, geographic fit and brand sentiment. Do not claim that an increase in mentions caused a change in enquiries without supporting attribution.
Crescendo’s take: share of voice needs a denominator and a buyer journey
We reject impressive-looking “AI visibility scores” that hide which prompts were tested and what counted as a mention. A modest gain on high-intent questions can be commercially more useful than a large gain on generic prompts. Publish your methodology, keep the sample stable and discuss leads as a separate metric.
For broader context on creating people-first source material, review Google Search Central’s content guidance. It is a useful publishing standard, not a guarantee of citations in AI systems.
Frequently asked questions
Is AI share of voice the same as SEO share of voice?
No. SEO share of voice estimates visibility across conventional search results and keyword sets. AI share of voice observes mentions or citations across sampled AI answers; the measurement systems and denominators are different.
Should I combine ChatGPT and Perplexity data?
Show each platform separately first. If you calculate an aggregate, disclose the weighting and sample design so one platform does not silently dominate.
Can an AI mention be counted without a link?
Yes, for mention rate. But it should not also be counted as a citation unless a verifiable source link appears.
Need a baseline for your brand? Explore our GEO services or contact Crescendo to discuss a buyer-intent prompt study.