What Is Share of Model, and How Is It Different From Share of Voice?
Share of voice measures a brand's share of attention in a channel. Depending on who measures it, that means advertising, media coverage, social conversation or search results. Share of Model measures how often, how prominently and how favourably AI models name a brand when people ask about its category. The agency Jellyfish launched a platform under that name in 2024, and the idea spread quickly. The difference is the source. Share of voice counts what brands and publishers put into the world. Share of Model looks at what AI systems say back after absorbing and retrieving that material. Both are useful, but they are not interchangeable, and Share of Model needs a precise definition before anyone reports it.
How do the two compare?
| Share of voice | Share of Model | |
|---|---|---|
| Question it answers | How much of the conversation or visibility in a channel do we have? | How often do AI models name us, and how, when asked about our category? |
| Data | Ad impressions, media mentions, social mentions, search rankings | AI answers to a defined set of prompts, run repeatedly |
| Influenced by | Spend, PR, content, SEO | Training data, retrieved web sources, reviews, mentions across the web |
| Typical formula | Our mentions ÷ all category mentions | Our mentions ÷ all brand mentions across the prompt set, or our mention rate |
| Stability | Fairly stable month to month | Varies between runs and with model updates |
| Includes description quality? | Sometimes, through sentiment | Often: position, sentiment and attributes |
In AI search, "AI share of voice" and Share of Model are often used for the same calculation. When they differ, Share of Model usually puts more weight on how the model describes the brand, and sometimes on answers without web search. Those answers reflect what the model itself has learned. Always ask which is meant.
How is Share of Model calculated?
A clear definition has four parts:
- The prompt set: a fixed list of category, problem and comparison questions, mostly unbranded.
- The engines and settings: which models or assistants, with web search on or off, and in which locations and languages.
- The runs: how many times each prompt is run per period.
- The formula: for example, your brand mentions divided by all mentions of tracked brands across all runs. Alternatively, report your mention rate: the share of runs that name you.
Many teams also record position and sentiment for each mention. Report the share, the mention rate and the competitor set together. A share rises if a competitor disappears from answers, even when your own visibility has not changed. Related metrics are covered in which AI visibility KPIs brands should track.
Are GEO and Share of Model useful KPIs for AI visibility?
Yes, as leading indicators, if they are defined carefully and read with their limits:
- Useful because they show whether AI systems include you when buyers ask, which matters before any click happens.
- Useful because they can be compared with competitors and tracked over time.
- Limited because results vary by prompt wording, engine and run, so small changes are often noise.
- Limited because they do not measure revenue. Pair them with AI referral traffic, self-reported attribution and pipeline.
- Limited because a share figure depends on the competitor set, and the competitor set can change.
Use Share of Model to guide content, PR and review work, and judge business impact with other measures.
How can agencies measure Share of Model instead of organic sessions?
Agencies should measure it alongside organic sessions, not instead of them. Sessions still matter for sites that depend on search traffic. To add Share of Model:
- Agree the prompt set with the client, split into unbranded and branded groups.
- Run each prompt several times per engine, with settings recorded.
- Report share, mention rate and position, with the number of runs behind each figure.
- Track description accuracy, since a mention with wrong facts can do harm.
- Explain the relationship: organic sessions show traffic from search, while Share of Model shows presence in answers that often produce no click.
A monthly report can show both on one page: sessions and conversions from search, and Share of Model with competitor comparison.
Share of model versus share of voice reporting for GEO agency contracts
Contracts should define exactly what is measured and avoid guarantees:
- Define the metric in the contract or its appendix: prompt set, engines, settings, runs, formula and competitor set.
- Record the baseline before work starts, measured the same way.
- State who owns the prompt set and data, and how changes to the prompt set are agreed.
- Do not guarantee a share target. No agency controls model outputs. Tie fees to work delivered and reporting, with Share of Model as a tracked outcome.
- Separate it from share of voice if both are reported, since they come from different data and move for different reasons.
- Note model changes, since a provider's update can move share figures regardless of the agency's work.
Does high Google visibility guarantee a high Share of Model?
No. Search visibility helps, but it is not the same thing:
- Google's own AI features draw widely. Ahrefs' 2026 analysis found only about 38% of AI Overview citations came from the top 10 organic results.
- Other assistants use other indexes. Claude's web search uses Brave Search. One analysis of about 35,000 Claude citations found 79.2% came from Brave's top 10, compared with 34% from Google's top 10.
- Answers without web search reflect what the model learned from mentions across the web, including reviews, articles and forums, not rankings.
- Mentions matter. Ahrefs' study of 75,000 brands found branded web mentions correlated far more strongly with AI Overview visibility than backlinks did.
A brand with strong rankings but few third-party mentions can have a low Share of Model. A brand widely discussed in reviews and media can score well despite modest rankings.
What might an illustrative report show?
Consider an illustrative running shoe brand tracking 50 prompts three times a month in four assistants. Its share of voice in search is strong, second in its category for organic visibility. Its Share of Model is 11%, fourth among six tracked brands. Answers without web search rarely name it. Answers with web search name it more often, citing its own pages. Two competitors with lower search visibility score higher, and their mentions come from running publications and review roundups. The agency's plan shifts budget from on-site content to product reviews in specialist publications, and tracks Share of Model quarterly alongside organic sessions.
What do people ask about Share of Model?
Is Share of Model the same as AI share of voice?
Often the same calculation under a different name. Check the definition, especially whether answers without web search and description quality are included.
What is a good Share of Model?
It depends on the number of competitors and the category. Compare your trend and your position against named competitors rather than an absolute target.
How often should it be measured?
Monthly for most brands, with enough runs per prompt to smooth out variation.
Should we measure answers with web search off?
It is worth a small sample. Answers without web search show what the model has absorbed about your brand, which changes slowly and reflects your long-term reputation across the web.
Can Share of Model be improved quickly?
Search-based answers can change within weeks. Answers from model knowledge change mainly with new model versions, so expect slower movement there.
Published by The Kovmere Team · updated Sep 29, 2026 · More resources