AI share of voice: how to measure it
Two numbers get mixed up all the time: visibility and share of voice. Here is the correct formula for each, how many answers you need, how to read the 95% range, and real ranges from the kennt.ai Visibility Index.
Visibility = answers that name your brand ÷ all answers. Share of voice = your brand's mentions ÷ all mentions of the tracked brands. Report both per week, engine and market, always with n and a 95% Wilson range; a change counts only when the ranges of two full weeks don't overlap.
Visibility vs. share of voice: two different numbers
Visibility asks: in how many answers are you named at all? Share of voice asks: of all the brand mentions in those answers, how many are yours? Both come from the same answers to unbranded buyer questions, but they answer different questions.
visibility = answers naming the brand ÷ all answers
share of voice = mentions of the brand ÷ mentions of all tracked brandsWorked example
100 answers, your brand is named in 40 of them, and the answers name the tracked brands 250 times in total. Visibility = 40 ÷ 100 = 40%. Share of voice = 40 ÷ 250 = 16%.
Both are right. Share of voice is lower because every answer names several brands. It also depends on which competitors you track: add one brand and everyone else's share drops. That is why visibility is the better headline and share of voice the better competitive comparison, always with the same set of brands.
How many prompts and answers you need
A percentage without n is a guess. The table shows the 95% Wilson range for a value of 50%, the widest case, at different numbers of answers:
| Answers (n) | 95% range at 50% | Uncertainty |
|---|---|---|
| 30 | 33–67% | ± 17 points |
| 50 | 37–63% | ± 13 points |
| 100 | 40–60% | ± 10 points |
| 200 | 43–57% | ± 7 points |
| 400 | 45–55% | ± 5 points |
| 1,000 | 47–53% | ± 3 points |
Computed with the Wilson formula below. At values further from 50% the range is a little narrower.
Practical rule: start with 30 or more unbranded prompts per market. With four engines that is enough answers for a usable weekly range per market; per engine, ranges stay wide until you have around 100 answers a week.
Answers to the same prompt are related: if ChatGPT names you for a question today, it likely does tomorrow. So 100 answers to 10 prompts carry less information than 100 answers to 100 prompts. kennt.ai uses a cluster-robust effective n for that reason, which is why its n is smaller than the raw answer count (methodology).
The 95% range (Wilson), worked through
The Wilson score interval works well for shares close to 0% or 100% and for small n, where the simple ± formula breaks down. With z = 1.96 for 95%:
p̂ = k ÷ n
centre = (p̂ + z²÷2n) ÷ (1 + z²÷n)
half width = z × √(p̂(1−p̂)÷n + z²÷4n²) ÷ (1 + z²÷n)
range = centre ± half widthExample: named in 30 of 100 answers, so p̂ = 30%. The denominator is 1.038, the centre 0.307, and the 95% range runs from 21.9% to 39.6%: about ± 9 points. With four times as many answers (400) at the same share, the range shrinks to about ± 4 points. Halving the uncertainty takes four times the answers.
Why daily values are noise
AI answers vary from one request to the next, even for the same prompt. With 40 answers a day, a value of 50% has a 95% range of about ± 15 points, so a jump of ten points from Monday to Tuesday says nothing. A week of the same volume (280 answers) gets you to about ± 6 points.
Rule that holds up: report weekly values, compare whole weeks, and call a change real only after two full weeks with non-overlapping 95% ranges. Never report a single day or a single position change as a trend.
Real ranges from the AI Visibility Index
The kennt.ai Visibility Index measures the same buyer questions every week for up to 15 brands per industry on ChatGPT, Gemini, Google AI Overviews and AI Mode. Below: the leader of 5 German industries (Edition week 46, 9 – 15 Nov 2026). Full rankings in the Index.
| Industry | Leader | Visibility | 95% range | Share of voice | n | Lead over no. 2 |
|---|---|---|---|---|---|---|
| CRM software | HubSpot | 87% | 77–93% | 23% | 69 | Clear (ranges don't overlap) |
| Online brokers | Scalable Capital | 74% | 64–81% | 20% | 102 | Not clear (ranges overlap with Trade Republic) |
| Accounting & tax | Lexware Office | 71% | 59–81% | 19% | 66 | Not clear (ranges overlap with sevdesk) |
| Neobanks | N26 | 46% | 37–56% | 16% | 95 | Not clear (ranges overlap with C24 Bank) |
| E-bikes | Bosch eBike Systems | 60% | 49–70% | 31% | 77 | Clear (ranges don't overlap) |
AI Visibility Index (Edition week 46, 9 – 15 Nov 2026). Non-branded answers; n = cluster-robust effective number of answers behind the leader's value. None of the brands is a kennt.ai customer.
Read it like this: the leaders are named in a large share of answers, yet their share of voice is only 16% to 31%, because each answer names several brands. And where the ranges of no. 1 and no. 2 overlap, the data cannot say who is ahead this week, even when the point values look far apart.
Which tool shows share of voice in ChatGPT, Gemini and Perplexity compared to competitors?
Most AI visibility trackers show some form of share of voice next to competitors. Before you trust the number, check four things:
- What is counted. Mentions ÷ all tracked brands' mentions (share of voice) or answers naming the brand (visibility)? Some tools call either one "share of voice".
- Per engine. ChatGPT, Gemini and Perplexity often disagree. A blended number hides that.
- n and a range. Every value should show how many answers are behind it and a 95% range.
- Which engines your plan includes. Perplexity is an add-on or a higher plan at several vendors; see what AI visibility tracking costs.
kennt.ai shows visibility and share of voice per week, engine and market with n and a 95% range, next to every tracked competitor. Free covers ChatGPT, Gemini, Google AI Overviews and AI Mode; Perplexity and Claude come with a custom plan from brutal.ai.
kennt.ai Free collects the answers, counts mentions, and reports visibility and share of voice per week with n and a 95% range, for €0. Or start with the free AI visibility check and see who AI names instead of you.
Questions
What is AI share of voice?
Your brand's mentions as a share of all mentions of the tracked brands in AI answers to unbranded buyer questions, per week, engine and market. See the glossary entry.
What is the difference between AI visibility and share of voice?
Visibility counts answers (in how many answers are you named?); share of voice counts mentions against competitors (of all mentions, how many are yours?). A brand can be named in most answers and still have a modest share of voice.
Which tool shows share of voice in ChatGPT, Gemini and Perplexity compared to competitors?
Most trackers do; check that the number is per engine, shows n and a 95% range, and that your plan includes Perplexity. kennt.ai shows both metrics per engine with n and ranges; Perplexity there needs a custom plan. Prices and engines per tool: cost comparison.
How many answers do I need for a reliable share of voice?
For a weekly value per market, a few hundred answers give a range of about ± 5 to 7 points. Start with 30 or more unbranded prompts per market and report weekly.
Can I compare share of voice between tools?
Only with the same prompts, engines, markets and set of tracked brands. Add or remove one competitor and every share changes.
Does AI recommend you?
The free AI visibility check asks ChatGPT, Gemini and Google AI Mode 10 buyer questions each, live. 45 seconds, no sign-up.