How to Measure Share of Voice in AI Search
Mention Rate Is Only the Starting Point
A simple GEO dashboard often begins with mention rate: the percentage of tracked AI answers that mention the brand.
That is useful, but it does not show the competitive picture. A brand can improve its mention rate while a competitor grows faster. It can also appear frequently as a minor alternative while another brand owns the primary recommendation.
AI search share of voice adds competitive context.
Define the Query Set First
Share of voice is only meaningful when the query set is controlled.
Build a stable group of questions across:
- category discovery
- use-case recommendations
- budget or specification constraints
- competitor comparisons
- alternatives
- brand fact validation
Basic Mention Share
The simplest calculation is:
Brand mentions divided by total mentions of all tracked brands.
If five brands receive 100 combined mentions and your brand receives 18, your mention share is 18 percent.
This is better than raw mention rate for competitive comparison, but it still treats every mention equally.
Weight by Recommendation Depth
AI answers often distinguish between primary recommendations, secondary options, and brief references.
A practical weighting model might assign:
- primary recommendation: 3 points
- secondary recommendation: 2 points
- brief mention: 1 point
- not mentioned: 0 points
The exact weights matter less than using the same rules consistently.
Include Position When Available
Some answers present ordered lists. First position usually carries more attention than fifth position.
You can add a position factor, but avoid creating a formula so complex that nobody trusts it. A simple score that teams understand is more useful than a mathematically elegant black box.
Segment the Result
A single overall percentage can hide important differences. Segment share of voice by:
- AI platform
- market and language
- query type
- product category
- user persona
- decision stage
Track Competitor Co-Mentions
Record which brands appear together. Co-mention data shows the competitive set AI has learned.
If your brand is repeatedly compared with a budget competitor while your strategy is premium, the problem is not only visibility. It is positioning.
Add Accuracy and Sentiment
Visibility is not automatically positive.
For each mention, track whether the description is accurate and whether the recommendation contains a caveat. A high share of voice built on outdated facts or negative warnings is not a win.
A useful report can show three layers:
- visibility share
- recommendation depth
- accurate positive share
Sampling Matters
AI answers can vary between runs. Use repeated samples for high-value queries and avoid drawing conclusions from one answer.
Document model, date, platform, region, language, and whether web search was enabled. Changes in model behavior can affect the result independently of your GEO work.
Turn the Metric Into Action
Share of voice should answer where the brand is losing and why.
Examples:
- low category share: strengthen broad category authority
- low use-case share: build scenario pages and proof
- strong mentions but weak depth: improve comparison and differentiation content
- high visibility with factual errors: fix source-of-truth pages
- loss on one platform: study the sources that platform cites
Bottom Line
AI search share of voice is not one universal industry number. It is a controlled competitive metric built from your own query set and consistent scoring rules.
Used well, it shows not only whether AI mentions the brand, but whether the brand is gaining a meaningful position in the recommendation landscape.