What is share of model?
Share of model is the proportion of brand mentions in AI model answers that belong to one brand, within a defined category. If a running shoe brand is named 30 times across a set of ChatGPT and Gemini answers to "best running shoes for flat feet" style questions, and all brands together are named 200 times, that brand's share of model is 15%.
Tom Roach defined it in Marketing Week as "the number of mentions of a brand by one or multiple LLMs, as a proportion of total mentions of brands in the same category" [1]. The idea is simple: when people ask an AI assistant for a recommendation, the brands the model names are the shortlist. Share of model measures how much of that shortlist a brand owns.
Share of model counts mentions, not links. A brand can score well on share of model even when the AI answer cites no source at all, because the model may name the brand from what it learned in training.
Who coined share of model?
Share of model was coined by Jack Smyth at the marketing agency Jellyfish, who started using the term in early 2024, according to his colleague Tom Roach [1]. Roach set out the concept in print in Marketing Week on July 11, 2024, and Jellyfish launched a commercial Share of Model platform on December 4, 2024 [1] [2].
2024: the term is coined
Roach placed share of model as the latest in a line of "share of" metrics, each tied to the dominant marketing technology of its time: share of market, then share of voice for media spend, then share of search in 2020, and now share of model for generative AI [1].
December 2024: a platform launches
Jellyfish launched its Share of Model platform on December 4, 2024, to track how ChatGPT, Google's Gemini and Meta's Llama perceive brands, products and services [2]. Jellyfish presents the name with a trademark symbol, so "Share of Model" is both a general term and a product name.
2025: the term reaches the mainstream
In June 2025, Harvard Business Review published a piece by David Dubois, John Dawson and Akansh Jaiswal that used share of model as its central measure, assessed through three dimensions: mention rate, the gap between human awareness and AI awareness of a brand, and brand and category sentiment [3].
How do you measure share of model?
To measure share of model, run a fixed set of category prompts through one or more AI assistants many times, count how often each brand is named, and divide each brand's mentions by the total brand mentions. The steps:
- Write prompts a real buyer would ask in your category, such as "what is the best accounting software for a small agency".
- Run each prompt repeatedly in each assistant you care about.
- Record every brand named in every answer.
- Divide your brand's mentions by all brand mentions.
- Repeat on a schedule and track the trend, not one reading.
The repetition matters more than most guides admit. In a study run by Rand Fishkin of SparkToro and Patrick O'Donnell of Gumshoe.ai, 600 volunteers ran 12 identical prompts through ChatGPT, Claude and Google's AI nearly 3,000 times. The odds of getting the same list of brands twice were under 1 in 100, and the odds of the same list in the same order were closer to 1 in 1,000 [4]. Some brands still appeared in 60% to 90% of runs [4]. The lesson: rank position in an AI answer is close to noise, but the share of runs a brand appears in is a usable signal.
Should you set temperature to zero when measuring share of model?
No, not if the goal is to measure what customers see. Some guides recommend setting the model's randomness (temperature) to zero so answers stay consistent between checks [5]. That only works through an API, and it measures a version of the model that no customer uses. People asking ChatGPT, Gemini or Google's AI features get answers with normal variation, which is exactly why the same prompt returns different brands [4].
A zero-temperature reading gives a stable number that may be stably wrong. A better approach is to keep normal settings, run each prompt enough times to see the spread, and report share of model as a range or an average over many runs.
What is the difference between share of model and share of voice?
Share of model measures how often AI models name a brand, while share of voice traditionally measures a brand's share of advertising or media presence in its category. Share of voice is something a brand can buy more of directly. Share of model is earned indirectly, through how the brand is represented in the content models learn from and retrieve.
| Metric | What it counts | Where |
|---|---|---|
| Share of voice | Brand's share of ad spend or media mentions | Paid and earned media |
| Share of search | Brand's share of branded search volume in a category | Search engines |
| Share of model | Brand's share of mentions in AI answers | AI assistants and AI search |
| Citation share of voice | Domain's share of sources cited in AI answers | AI answers that show sources |
What is the difference between share of model and citation share of voice?
Share of model counts brand names in the answer text, while citation share of voice counts which websites an AI answer cites as sources. The two often diverge. A software review site can be cited constantly while recommending competitors, giving it high citation share and the competitors high share of model. A well-known brand can be named in an answer that cites no source at all.
Track both if AI search matters to the business. Share of model tells you whether buyers hear your name. Citation share tells you whether your pages are the evidence.
What are the limits of share of model?
The main limit of share of model is that no standard method exists, so two share of model figures from different tools or agencies are rarely comparable. Each vendor chooses its own prompts, assistants, number of runs and counting rules. Changing any one of those changes the number.
Other limits to keep in mind:
- Prompt choice drives the result. A prompt set written around a brand's strengths will flatter it. There is no public data on what people actually type into AI assistants at scale, so prompt sets are educated guesses.
- Mentions are not endorsements. A brand named as "the expensive option to avoid" counts as a mention. HBR's framework adds sentiment for this reason [3].
- Models change. A model update can move share of model overnight with no change in the brand's own activity.
- Sample size. With answer lists this unstable, a handful of runs per prompt can produce swings that are pure chance [4].
No published research yet links a given share of model to sales or market share. The metric is plausible as a leading indicator, but that link is a claim, not an established finding.
How do you improve share of model?
You improve share of model by making the brand easy for AI systems to associate with the category, through consistent, specific coverage on the pages models learn from and retrieve. In practice that means being named in independent reviews, comparisons, lists and discussions for the category, keeping the brand's own pages clear about what it does and for whom, and checking for brand hallucinations where a model gets facts about the brand wrong.
Measure before and after any change, over many runs, and judge the trend rather than any single answer.