Quick answer
Useful AI visibility KPIs are mention rate, recommendation rate, citation rate, share of voice, average position and accuracy of key facts. Each needs a defined denominator, a fixed prompt set, a platform breakdown and a stated sample size, otherwise the number cannot be compared over time.
Key takeaways
- Every KPI needs a numerator, a denominator and a sample size.
- Report by platform, topic and location, not only as one blended figure.
- Add an accuracy metric, because being described wrongly can be worse than being absent.
| KPI | Definition | Watch out for |
|---|---|---|
| Mention rate | Answers that name the brand, out of eligible answers | Changing the prompt set between periods |
| Recommendation rate | Answers that actively recommend the brand | Confusing a mention with a recommendation |
| Citation rate | Answers that cite a brand owned page | Ignoring third party citations |
| Share of voice | Brand mentions as a share of all tracked brand mentions | An incomplete competitor list |
| Average position | Where the brand is listed when several are named | Treating it as a fixed rank |
| Fact accuracy | Key facts stated correctly, out of facts checked | Checking facts once, not repeatedly |
How should you report them?
- Show the sample size next to every rate, for example 12 of 40.
- Report changes only when the prompt set and platforms are unchanged.
- Keep a link from every figure to the answers behind it.
- Separate measured rates from interpreted labels such as sentiment.
Definitions you can put in a report
| KPI | Numerator | Denominator |
|---|---|---|
| Mention rate | Answers that name the brand | Eligible completed answers |
| Recommendation rate | Answers that actively recommend the brand | Eligible completed answers |
| Citation rate | Answers that show a page of the brand as a source | Eligible answers on platforms that show sources |
| Share of voice | Mentions of the brand | All mentions of tracked brands in the same answers |
| Fact accuracy | Checked facts stated correctly | Facts checked |
Eligibility deserves a written rule. A common rule: an answer is eligible if the run completed and the prompt is one the brand could reasonably appear in. Write the rule down once and apply it every time.
Segment before you summarise
A single blended figure hides the information you need. Split every KPI by platform, by topic and by intent, and by language and location where relevant. The headline can then combine them, with the breakdown one click away. When a summary number falls, the segments tell you where.
Set targets that are honest about noise
- Measure your baseline, then repeat it once without changes to learn the normal variation.
- Set targets larger than that variation, so you are not chasing noise.
- Set targets on the segments that matter commercially, not on the easiest ones.
- Review targets quarterly, since platforms change.
Pair the KPIs with business measures
- AI referral sessions in analytics, treated as a lower bound.
- Branded search and direct traffic trends, as supporting evidence only.
- Enquiry or sales attribution where customers say they used an assistant.
- Coverage of the topics that carry the most revenue.
The most common reporting error
Reporting rhythm and ownership
Give each KPI an owner who understands how it is calculated and can explain a change. Agree the review rhythm in advance: a light monthly look at the headline rates and a fuller quarterly review of segments, sources and targets. Regular, predictable reporting builds more trust than an impressive report that appears once.
When a KPI moves, resist the urge to explain it before checking it. Confirm that the prompt set, platforms and number of runs match the previous period, read a sample of the answers behind the change, and only then write the explanation. A short, correct explanation is worth more than a long, confident one.
Is there one AI visibility score?
A single blended score hides differences between platforms and topics. Report the underlying rates, and use a composite only alongside them.
How many runs per prompt do I need?
Enough that the result is stable for the decision you are making. Report the sample size so readers can judge it.
How often should we measure?
On a regular schedule that fits how fast your content and the platforms change, and after any major change you make.
Is a composite AI visibility score useful?
Only if it is defined clearly and shown next to the rates it combines. On its own, a composite hides the differences between platforms and topics that you most need to see.
How do I choose targets?
Baseline first, learn the noise, then set targets beyond it on the segments that matter most to the business.
Cite this page
BrandWater AI Research. (2026, 11 September 2026). AI visibility KPIs: the metrics worth reporting, and how to define them. https://brandwaterai.in/guides/ai-visibility-kpis
How we work
Figures are dated and linked to their sources. Where none exist we say so. Read our methodology and AI transparency pages.