Definition
Share of model is the percentage of brand mentions in AI-generated answers that belong to one brand, out of all mentions of brands in the same category, across a defined set of prompts. If ChatGPT names some project management tool 400 times across your tracked prompts and 100 of those mentions are yours, your share of model in ChatGPT is 25%.
The term came from the agency Jellyfish, where Jack Smyth began using it in early 2024. Tom Roach, also at Jellyfish, defined it in Marketing Week as the number of mentions of a brand by one or more LLMs as a proportion of total mentions of brands in the same category. It deliberately echoes share of voice (a brand's share of advertising or media exposure) and share of search (a brand's share of branded searches in its category), both of which marketers use as leading indicators of market share.
In practice, share of model and AI share of voice are calculated the same way. The useful distinction is scope. Share of model emphasizes the per-model view: a brand can hold 40% of mentions in one engine and 5% in another, because each engine retrieves different sources, runs different searches and was trained differently. AI share of voice is often reported as a blended figure across engines. Promptwatch reports this metric as Share of Voice and lets you filter it to a single model, which gives you share of model directly.
A few measurement choices decide what the number means:
- The prompt set. Share of model only describes the prompts you track. A set skewed toward informational questions will show different results from one built around buying decisions.
- The competitor set. Your share is relative to the brands in the denominator. Adding a competitor lowers everyone's share; comparing against every brand detected in answers lowers it further.
- Counting rules. Counting a brand once per response, however often its name appears, stops one enthusiastic answer from inflating the figure and makes the number read as 'share of answers we appear in'.
- Collection method. Answers from consumer apps can differ from API answers for the same model, so be consistent and prefer the interface your customers use. See UI vs API tracking.
- Sample size and timing. AI answers vary between runs, so trends over weeks are more reliable than a single day's reading.
Share of model is a presence metric. It does not say if you were the headline recommendation or a footnote, or if the answer was positive. Pair it with a prominence metric such as an AI visibility score and with sentiment monitoring. Your absolute mention count can rise while your share falls, because a competitor grew faster; that is the signal the metric exists to catch.
To move it, start with prompts where competitors are mentioned and you are not, check which sources those answers cite, and work on presence in those sources: reviews, comparison articles, forums and your own pages that answer the question directly. Share of model cannot be bought with ads in the organic answer, though ads in AI products (such as ChatGPT ads and ads in Google's AI features) are a separate, paid channel. Our explainer on AI share of voice covers the calculation in more detail. Related: brand inclusion rate, mention rate and citation share.
Examples of Share of Model
- A CRM vendor tracks 150 prompts and finds a 31% share of model in ChatGPT but 9% in Perplexity. The Perplexity answers cite G2 category pages and two comparison blogs where the vendor is missing, which sets the outreach list for the quarter.
- A running shoe brand's mention count in AI Overviews rises 20% over six months while its share of model drops from 18% to 14%, because a newer competitor's mentions tripled in the same period.
- An agency reports share of model to a client both against five named competitors (27%) and against every brand detected in answers (11%), and explains that the second number includes retailers and marketplaces the client does not compete with directly.
- A fintech compares share of model before and after a model release and sees a five-point drop in one engine with no change in cited sources, so it waits two weeks for the trend to settle before changing its content plan.
