What Share of Model is
When a shopper asks ChatGPT or Gemini “which robot vacuum should I buy”, the model usually names several brands. Share of Model answers the question of how often you are among them and how often your competitors are. It is the equivalent of share of voice, except that instead of ad impressions or media mentions it counts mentions in AI answers.
The metric grew out of GEO and LLMO: when an answer closes the question without a visit to any site, as in zero-click search, the fact that the brand was named is what matters.
Where the term comes from
The term is promoted by the agency Jellyfish, part of The Brandtech Group. On 4 December 2024 it launched the Share of Model™ platform: it analyses how different LLMs — the release names ChatGPT, Gemini and Llama — perceive brands, products and services, how the models see competitors, and which messaging and content should be improved. Share of Model™ is a Jellyfish trademark, but in the industry the phrase is also used as the generic name of the metric.
How to measure it
| Parameter | What to record |
|---|---|
| Prompt set | 30–100 non-branded category prompts: choosing, comparing, buying on a budget |
| Models | each assistant separately, with and without web search |
| Repeats | 3–5 runs per prompt: answers are non-deterministic |
| Mention | whether the brand appears and which competitors sit next to it |
| Position | where the brand sits in the list of recommendations |
| Sentiment | whether the model recommends, stays neutral or warns against |
SoM = brand mentions / mentions of all brands in the category × 100%
Example: 120 brand mentions out of 600 mentions of all brands → SoM = 20%
Share of Model, share of voice and citability
| Metric | What it counts | Where |
|---|---|---|
| Share of voice | the brand’s share of advertising and mentions | media, advertising, search results |
| AI citability | links to your content as a source | AI answers |
| Share of Model | brand mentions, with or without a link | AI answers |
Important: high citability does not guarantee a high share of model. A model can take facts from your buying guide and still recommend your competitors’ brands.
How to use it: checklist and mistakes
- Start with a baseline of 30 prompts in the two or three models your audience actually uses, and repeat it monthly.
- Split the prompts by stage of choice: general, comparative, budget-bound, task-specific. This shows at which stage the brand drops out of the answers.
- For prompts where the brand is missing, check which sources the model cites and work on those: guides, reviews, ratings, AI Overviews on the same topics.
- Mistake: a single run. One answer from a model is a random sample, not a metric.
- Mistake: branded prompts. A prompt like “what is brand N” shows what the model knows about you, not your share of the category.
- Mistake: mixing modes. Count answers with and without web search separately.