How to measure AI share of voice for an enterprise brand
A working definition, the four metrics teams confuse, and a worked example across engines, markets and languages.
When a buyer asks an assistant "which developer should I choose in the UAE?", there is no page two. There's one answer, which usually names two to four brands. Share of answer asks a simple question: out of every relevant answer, how many include you?
AI share of voice is now a standard GEO metric, but teams define it in incompatible ways and then argue over the numbers. Here's how to fix the definition before you optimise anything.
Four metrics that get mixed up
Each of these measures something different. A mature dashboard tracks all four for a fixed list of competitors[1][2].
| Metric | What it counts | Use it for |
|---|---|---|
| Mention rate | Share of answers that name your brand at all | Baseline visibility |
| Citation rate | Share of answers that link or footnote your domain | Owned-content influence |
| Recommendation rate | Share of answers where you are the suggested choice, not just listed | Commercial impact |
| Share of voice | Your mentions ÷ all brand mentions (you + tracked competitors) | Competitive position |
| Fact accuracy | Share of mentions where key facts are correct | Brand and legal risk |
Build the prompt set before the dashboard
Engines answer questions, not keywords, so the unit is a prompt library: realistic buyer questions grouped by intent. For one market, 15–50 prompts across brand, category and comparison clusters is a workable start[2]. At enterprise scale, multiply by markets and languages.
A worked example
An illustrative UAE developer brand, 200 prompts per week, run on five surfaces. The spread between engines is normal. One published example shows a single brand at 22% on Perplexity, 18% on ChatGPT and 6% on Gemini in the same week[1]. That's why a blended average hides the real work.
Cadence and hygiene
- Weekly. Daily sampling is noisy and monthly loses signal[1].
- Freeze the competitor set. Share of voice is only comparable if the denominator stays the same.
- Log the answer, not just the score. Keep the full text and the citations behind every data point so a change can be traced.
- Split by language. Arabic and English answers for the same market often disagree about who leads.
