AI share of voice measures your presence in AI-generated answers relative to competitors. You pick a set of prompts that matter to your business, run them through one or more AI tools, and count how often each brand is mentioned or cited. The result is a share: of all the brand appearances in those answers, what portion was yours.
How to calculate it
The simplest version counts answers that mention each brand. If you run 50 prompts and your brand is named in some of them, your share is your mentions divided by the total mentions of every tracked brand. Some teams weight by position, giving more credit to a brand named first or recommended outright, and some track citations as a separate share.
Whatever method you use, keep it fixed. The number is only useful as a trend and as a comparison with competitors measured the same way.
Building the prompt set
The prompt set decides what the metric means. Base it on how your buyers actually search and ask, not on what flatters you.
- Category prompts: “best [category] software for [team type]”.
- Alternatives prompts: “alternatives to [competitor]”.
- Comparison prompts: “[you] vs [competitor]”.
- Problem prompts: “how do I [job your product does]”.
- Integration and requirement prompts: “[category] tool that integrates with [platform]”.
Limits of the metric
AI answers are not stable. The same prompt can give different answers to different users, on different days, with different models, and with or without memory or personalization. Logged-out testing does not match what a signed-in user with history sees. Treat AI share of voice as a directional sample, run each prompt more than once, and avoid reading meaning into small changes.
There is also no public data on how many people send any given prompt, so a high share on a prompt nobody asks is worth little. Weight your prompt set toward questions you hear from real buyers.
Example
A fictional project management tool
Tasklane, a made-up project tool for agencies, tracks 60 prompts across ChatGPT, Perplexity and Google AI Mode, three runs each, monthly. It records which of five competitors appear in each answer. Over a quarter, its share rises on integration prompts after it publishes clear integration pages, and stays flat on “best project management software”, where answers are dominated by large review sites. That tells the team where on-site work helps and where off-site work is needed.
Common mistakes
Too few prompts or runs
A handful of prompts run once is mostly noise. Use enough prompts to cover the buying journey and repeat runs.
Changing the prompt set constantly
Every change breaks the trend line. Add prompts deliberately and keep the core set fixed.
Reporting it as market share
It is a share of appearances in a sample you chose, not a share of buyers or revenue.
Questions
Several commercial tools track it, each with its own methods. Results from different tools are rarely comparable, so pick one approach and stick with it.
Track them separately. They move for different reasons and call for different work.
