Share of Voice (AI Answers)
A meta-metric that quantifies how often and how prominently a brand surfaces in AI answers relative to competitor brands. The industry standard is a hybrid SoV that combines four raw metrics: visibility share + mention share + citation share + position share
What is Share of Voice (AI Answers)?
Share of Voice (SoV · AI Answers) measures how often and how prominently a brand appears in AI answers (ChatGPT · Perplexity · Gemini, etc.) relative to competitor brands. Where classical SoV measured advertising impressions or media coverage share, AI-answer SoV measures the share of brand citation inside the LLM answers themselves.
Often reported as a single number, but in practice it is a hybrid metric combining several raw signals. Single-dimension measurement amplifies measurement error and bias.
Four raw metrics + hybrid SoV
The industry standard is to measure four raw signals separately, then combine them into a hybrid score.
| Raw metric | Abbr. | What it measures |
|---|---|---|
| Visibility Share | VS | Frequency of brand appearance in AI answers / total brand appearances |
| Citation Share | CS | Share of source URLs in answers that point to the brand's domain / total source URLs |
| Position Share | PS | Weighted score for where the brand appears in the answer (first paragraph > middle > last) |
| Mention Share | MS | Share of brand-name occurrences in answer text |
Each raw metric carries independent meaning. Example — high VS with low PS means the brand is appearing but only at the end of answers. High CS signals strong authority.
Hybrid SoV combines the four with tool-specific weights. Weight distributions vary by tool but typically: Visibility (35–45%) + Citation (25%) + Position (20–25%) + Sentiment (10–15%). Useful as a single number for executive reporting, but the four raw metrics must accompany it for root-cause analysis.
Measurement pitfalls
(1) Both-absent cases. If neither the brand nor competitors appear in an answer, that response should be excluded from the denominator. Including it produces SoV = 0/0 = 0, polluting the data with meaningless points.
(2) Untracked brand handling. Including untracked brands (brands not explicitly named by the user) in the denominator destabilizes time-series. Industry standard: denominator = user-specified brands only.
(3) Platform normalization. Response length and citation frequency distribute differently across LLMs. Simple averaging lets verbose LLMs dominate the result. The correct pattern: group by LLM → compute per-LLM share → take the arithmetic mean.
(4) Tracked-brand coverage gate. When the share of responses where no tracked brand appears exceeds a threshold (say 80%), measurement reliability itself collapses. A reliable tool surfaces this gate explicitly.
How market tools differ on SoV
| Tool | SoV approach | Competitor cap |
|---|---|---|
| Profound | Hybrid SoV (weighted combination) | User-defined |
| Otterly AI | Brand Visibility Index (single metric) | 4–5 |
| Ahrefs Brand Radar | Brand mention share | User-defined |
| Semrush AI Visibility | Visibility share + sentiment separated | 9 |
| RanketAI | Hybrid SoV (four raw metrics + hybrid) | Multiple |
Because each tool defines and calculates SoV differently, you should not directly compare SoV values across tools. Same-tool time-series comparison is the valid pattern.
Frequently asked questions
Q. My SoV is up but revenue isn't moving?
SoV is a top-of-funnel awareness metric; revenue is a bottom-of-funnel conversion metric. Multiple stages (consideration · trial · purchase) sit between them. SoV gains do not immediately translate to revenue — a 3–6 month lag is common.
Q. Hybrid SoV or the four raw metrics — which matters more?
Depends on the use case. Executive reporting favors the single hybrid number. Action planning (which signal to improve) requires the raw four-way decomposition. Reporting both together is the standard.
Q. Can SoV values be compared across tools?
Not recommended. Tools differ on denominator definition (untracked brand inclusion), weights, and normalization. Only same-tool time-series comparison (e.g., January vs April) is meaningful.
Q. What's the minimum prompt count for reliable SoV?
Reliable measurement requires a Cartesian of multiple prompts × multiple LLMs. Statistically, N≥12 responses is the minimum for meaningful signal; N≥30 enables percentile analysis. Each tool's entry tier caps prompts differently — securing adequate prompt headroom is essential when selecting a tool.