tonecast
Free tool

AI visibility calculator

AI answers vary from one run to the next, so a visibility rate measured on a handful of answers is mostly noise: a brand named in 3 of 5 answers has a true rate anywhere between about 23% and 88%. This calculator gives the 95% confidence interval of a rate, the number of answers you need for a given precision, and whether a change between two periods is real.

Last updated · Written by the Tonecast team · Free, no account, nothing leaves your browser

What range is my visibility rate really in?

Count the answers you collected and how many of them name your brand (or recommend it, or cite your site: any yes/no outcome works). The interval is the Wilson score interval at 95%, which behaves well with small samples and with rates close to 0% or 100%.

Observed rate 60%. 95% interval: 23% to 88%.

How many AI answers do I need?

Choose the precision you want (the margin of error, in percentage points) and your best guess of the rate. If you have no idea, leave it at 50%: that is the worst case and gives the largest number.

You need about 97 answers per engine for ±10 points.

For readers without JavaScript, the same numbers at 95% confidence:

Margin of errorRate near 50%Rate near 20% or 80%
± 5 points385 answers246 answers
± 10 points97 answers62 answers
± 15 points43 answers28 answers
± 20 points25 answers16 answers

These counts are per engine, and per language or country if you report them separately: ChatGPT, Claude, Gemini and Perplexity answer differently, so a pooled number hides the differences you most need to see.

Did my AI visibility really change?

Enter the two periods. The calculator gives the difference and its 95% interval (Newcombe's method, built on two Wilson intervals). If the interval includes zero, the data cannot tell a real change from chance.

From 45% to 60%: +15 points, 95% interval −7 to +35 points. Not distinguishable from noise.

What these numbers assume. Each answer counts as an independent observation. Answers to the same prompt resemble each other more than answers to different prompts, so when a rate mixes several prompts the real uncertainty is somewhat larger than shown: prefer many prompts with a few runs each to one prompt run many times.

The interval covers sampling noise only. It says nothing about whether your prompts are the questions people really ask, or whether "names the brand" was judged correctly.

Frequently asked questions

How many times should I ask ChatGPT the same question to measure my brand's visibility?

It depends on the precision you need. For a margin of error of about 10 percentage points you need roughly 100 answers in the worst case (a rate near 50%), and about 385 for 5 points. With 5 or 10 answers, a visibility rate is compatible with a very wide range of true values.

What is a confidence interval for an AI visibility rate?

It is the range of true visibility rates that are compatible with what you observed. If a brand was named in 3 of 5 answers, the 95% Wilson interval runs from about 23% to 88%: the observed 60% says little on its own.

How do I know if my AI visibility really changed?

Compare the two periods with a confidence interval for the difference between the two rates. If the interval excludes zero, the change is larger than sampling noise at the 95% level; if it includes zero, the data cannot distinguish a real change from chance.

Tonecast applies the same statistics to every number it reports: each AI visibility rate, recommendation rate and Tone in AI score comes with its 95% interval, per engine, prompt and language. The methodology explains the rest.