How to measure AI Share of Voice
AI Share of Voice is how often AI names your business versus competitors when customers ask for the best in your category. Measure it by running a fixed set of real customer questions across ChatGPT, Perplexity, Gemini and Claude, logging every business named, and dividing your mentions by the total answers. Run it monthly, per engine, and watch the trend — not a single reading.
What is AI Share of Voice?
AI Share of Voice (SoV) is the share of AI answers that name your business when customers ask category questions. It is the GEO-era successor to a keyword ranking: instead of "where do I sit in the list of links," it asks "how often does the AI name me in the answer at all — and how prominently." If you run 20 prompts across 4 engines — 80 answers — and you're named in 12 of them, your AI Share of Voice is 15%.
The formula
| Metric | How to calculate it |
|---|---|
| AI Share of Voice | Answers naming you ÷ total answers run |
| Appearance / coverage rate | Distinct prompts you appear in ÷ total prompts |
| Per-engine SoV | The same, calculated separately for each of the 4 engines |
| Recommendation rank | Score each mention 1–4 (see scale below) and track the average |
Step 1 — Choose your prompt set
Pick 10–20 questions your customers actually ask — not keywords, full questions. A good set spans five shapes: one category ("best [X] in Singapore"), one need or specialty, one price ("affordable [X] Singapore"), one location ("best [X] near [area]"), and one decision or comparison ("how to choose an [X]"). Keep the set fixed so you can compare month to month.
Step 2 — Run each prompt across all four engines
Run every prompt on ChatGPT, Perplexity, Gemini and Claude. To keep readings clean and comparable:
- Fresh, logged-out (or temporary) chat per query — so your history doesn't bias the answer.
- Singapore location, no VPN — local results differ from a generic run.
- Web access on for ChatGPT and Claude.
- Log the date — answers change over time.
10 prompts × 4 engines = 40 answers; 20 prompts × 4 engines = 80. That's your denominator.
Step 3 — Log every business named
For each answer, record three things in a simple sheet: the businesses named, in order; your recommendation rank (the 1–4 scale below); and the sources the answer cited. That last column is gold — it tells you exactly which third-party pages you'd need to appear on to get named.
Step 4 — Score your recommendation rank
Being named at all is good; being the lead name is better. Score each appearance:
| Score | Meaning |
|---|---|
| 4 — Top pick | Named first / as the single recommendation |
| 3 — Featured | In the main shortlist with a positive descriptor |
| 2 — Passing mention | Listed, but without emphasis |
| 1 — Absent | Not named at all |
Step 5 — Calculate, by engine and overall
Count the answers that name you, divide by the total, and you have your overall AI Share of Voice. Then split it per engine — this matters, because the four rarely agree. In our own Singapore accounting & corporate-services study, one firm's share of voice ran from 25% on ChatGPT to 83% on Gemini. A single blended number would have hidden that completely.
Step 6 — Handle non-determinism: average and trend
The same prompt can return different businesses on different runs. So don't treat one snapshot as truth: run each prompt 2–3 times, average the result, and report the trend over months. Direction of travel — is your SoV rising? — is the real signal. (One-off audits should be clearly labelled single-run.)
Doing it manually vs. using a tool
The manual method above is free, rigorous, and the best way to learn what AI actually says about you — it's how we run our own studies. Once you want to track dozens of prompts every week, dedicated AI visibility tracking tools automate the querying and logging. We compare them in the best AI visibility tracking tools for 2026.
A measurement-honesty note
AI Share of Voice is real and trackable, but younger and messier than Google Analytics. There's no "Search Console for AI," answers are non-deterministic, and attribution is partly zero-click. The rigorous approach is a fixed prompt set, multiple runs, per-engine reporting, and a trend line — and being upfront that it's a directional measure, not a precise league table. Any provider claiming pinpoint precision is overselling.
Want your AI Share of Voice measured for you?
We'll run five of the highest-intent questions your customers ask, across ChatGPT, Perplexity, Gemini and Claude, and send a one-page snapshot of where you stand against competitors. Free, no call required.
Get your free 5-prompt check →