Count the answers.
Not the vibes.
Enter a batch of AI-answer observations. Get your recommendation rate, citation rate, AI share of voice, and a confidence range that shows how much uncertainty the sample still carries.
One batch in. Four honest numbers out.
The example below represents 90 answers: your business appeared in 18, competitors were named 72 times in total, and your website was cited in 9. Replace those counts with yours.
Your mentions ÷ all measured brand and competitor mentions.
95% range: 12.9%–29.4%
Answers that linked to your own domain.
Useful for direction. Repeat the same protocol before calling a small change real.
Do not have a clean observation set? The free Routeless Radar scan asks fixed customer questions across ChatGPT, Gemini, and Perplexity and saves the answer, source, model, and timestamp for each observation.
Collect the answers for meKeep the question set fixed.
Share of voice only becomes a trend when each period measures the same thing. Use the same customer questions, cities, assistants, account state, and number of observations. Start fresh conversations so earlier turns do not steer the answer, and record the model and date because both change.
Recommendation rate answers “how often did this batch name us?” AI share of voice answers “of every measured business mention, what share was ours?” Citation rate answers “how often did the answer use our own website as evidence?” They are related, but they are not interchangeable.
The confidence range is a Wilson binomial interval around recommendation rate. It is more honest than pretending 18 of 90 means the true rate is exactly 20%. It still assumes observations are independent; repeated prompts and shared retrieval sources weaken that assumption, so use the range as a warning label rather than a promise.
Frequently asked questions
What is AI share of voice?
AI share of voice is your business's mentions divided by all measured mentions of your business and its competitors. It describes how much of the observed recommendation set you own; it is not a Google ranking or a guarantee that the next answer will be the same.
How many AI answers should I sample?
Start with at least 30 independent answers for a directional baseline and aim for 100 or more before treating small movements as meaningful. Use the same fixed questions, locations, assistants, and collection method each time so periods remain comparable.
Why can competitor mentions exceed the number of answers?
One AI answer can recommend several competitors. Count each named business once per answer, so the combined competitor mention total can be greater than the number of answers sampled. Your own recommendation rate still cannot exceed 100%.
What does the confidence range mean?
The calculator uses a 95% Wilson interval around your recommendation rate. It shows how imprecise a finite sample is, but it does not correct for repeated prompts, changing models, personalization, or other dependencies in AI answers.