Measuring AI share of voice — the metric that matters most
Visibility score is the headline KPI. But Share of Voice is the metric that tells you if you're winning or losing the market. Here is how to define, calculate, and track it.
What share of voice means in AI search
Classical share of voice in advertising = your ad spend / total category spend. In AI search, it's analogous: your mentions / total brand mentions in your prompt set, across engines, in a window. A 30 % share of voice means three of every ten brand mentions in answers to your prompts go to you.
How to calculate it cleanly
- Define the prompt set. 30–200 buyer-relevant prompts. Stable across measurement periods.
- Define the competitor set. 5–15 brands you compete with. Include yourself.
- Run the prompt set across all relevant engines daily. ChatGPT, Perplexity, Gemini, Claude, Google AIO at minimum.
- Count brand mentions per response. A mention = the brand name appears in the answer text or as a cited source. De-duplicate per response (one mention per brand per response, even if mentioned twice).
- Divide. SoV = your mentions / total mentions across all tracked brands. Express as percent.
Don't include your own prompts (where your brand name is in the prompt itself) — those distort the number upward. Pure category prompts only.
What good looks like
There is no universal good number — it depends on category density. As a rough framework:
- Below 10 % — you are functionally invisible. Most users will not see you in any natural query.
- 10–25 % — you are a known option but not top-of-mind. Acceptable as a third or fourth-place player.
- 25–50 % — you are a frontrunner. Users see you in roughly half their queries.
- Above 50 % — you are dominant. This is rare and usually means a small competitor set or a niche category.
Watch the trend, not the level
A 30 % SoV that's been falling 1 % per month is a brand in trouble. A 15 % SoV that's been climbing 1 % per month is a brand on the rise. Trend over six months is a stronger signal than absolute level at one point in time.
Three failure modes that make SoV lie
1. Prompt set drift
If you keep adding new prompts to your set, your SoV will trend down purely because you're adding prompts you haven't optimised for yet. Lock the prompt set for at least 90-day measurement windows.
2. Competitor set staleness
If a new competitor enters your category and you don't add them to the set, your SoV will look stable while the actual category share dilutes. Audit your competitor set quarterly.
3. Engine mix changes
If you only track ChatGPT and one quarter Perplexity surges in usage, your reported SoV is no longer representative. Weight engine SoV by actual usage in your audience (proxied via your buyer-research surveys or analytics).
Beyond SoV: the metric stack
SoV is the headline. The full stack:
- Visibility Score — share of prompts where you appear (regardless of competitors).
- Share of Voice — your share of mentions vs. competitors.
- Citation Count — how often you're cited as a numbered source.
- Sentiment Mix — positive / neutral / negative tone across mentions.
- Recommendation Rate — how often models actively recommend you (vs. just listing you).
Tracking the full stack against a stable prompt set across all major engines, weekly, is the foundation of any serious GEO program.