# What is AI share of voice?

**Short answer:** AI share of voice (SoV) is the percentage of all brand mentions in AI-generated answers for your category that belong to your brand. If AI answers mention your brand 12 times and competitors are named 88 times across the same prompt set, your AI SoV is about 12%.

## How it is calculated

The honest formula is:

```
SoV = your brand mentions / total brand mentions in your category
```

Four decisions change the number, and any tool you compare must state them:

**Which prompts.** A set of 40 buyer questions produces a different distribution than a set of 400. Generic head terms favor incumbents; specific long-tail questions favor specialists.

**Which engines.** ChatGPT, Gemini, Perplexity, Copilot and Claude have different evidence sets and different retrieval behavior. A blended score can hide per-engine failure.

**Which answers count.** One answer usually names several brands. Counting every mention once per answer is the common convention; counting only the first recommendation produces a much harsher metric.

**Which time window.** AI answers change. A monthly average is stable enough to trend; a single day's sample is noise.

## Why AI SoV differs from web share of voice

Classic SoV is measured in search result positions and share of clicks. It assumes a ranked list. AI answers have no ranked list — they are generated prose that may compare four brands, recommend one, or mention none.

Three consequences:

1. **Presence beats position.** There is no "rank 3" inside an answer. You are either in the answer or not.
2. **Citations matter separately.** Being named without being cited sends no traffic. SoV measures mindshare; citation rate measures distribution.
3. **Answers are unstable.** The same prompt can return different brand sets between runs. Trends need enough samples to survive that variance.

## What a useful SoV report contains

- Per-engine breakdown, not only a blended figure.
- The prompt list, with per-prompt results.
- Raw answer text for every counted mention.
- A time series with sample size stated per point.
- Category definition — "brand mentions" needs an explicit list of brands counted as competitors, or the denominator is undefined.

Without the prompt list and the raw answers, an SoV number is unauditable.

## How to use AI SoV

**As a trend, not a score.** Track month over month. A single number is meaningless without a baseline and a direction.

**Diagnose by engine.** If you are strong in Gemini and absent in ChatGPT, the problem is likely retrieval access or content structure, not brand authority.

**Segment by question type.** Compare "what is X" questions against "best X for Y" questions. Definitions are won by content quality; recommendations are won by corroboration.

**Pair it with citations.** SoV up but citations flat means you are being named but not sourced. That is a conversion problem, not a visibility problem.

## Common mistakes

- Treating SoV as a KPI without a target category. Share of a category you invented is not a market position.
- Comparing your SoV against a competitor's dashboard. Different prompt sets produce incomparable numbers.
- Ignoring accuracy. A high SoV with wrong facts is worse than a low SoV with correct ones.
- Sampling once per month with four prompts and calling it a trend.

## Related reading

- [How to track AI referral traffic in GA4](/guides/track-ai-referral-traffic-ga4)
- [How do I know if AI cites my website?](/answers/how-do-i-know-if-ai-cites-my-website)
- [Best GEO tools](/compare/best-geo-tools)
- [GEO glossary](/glossary/geo-terms)