What is share of model?
Share of model describes how often AI assistants name a brand in answers to a fixed set of buying questions. It plays the role that ranking position plays in search, for a channel that answers with a few named products instead of a list of links.
An assistant asked “which waterproof cycling jacket should I buy under CHF 250” names two or three products, sometimes cites sources, and the person often acts without clicking. How often you are named, in which position and from which sources is what this metric counts. We use the term in lower case and descriptively, since vendors also use it as a product name.
How is share of AI answers calculated?
Two definitions are in use, and a report should say which one it gives:
| Definition | Formula | What it tells you |
|---|---|---|
| Presence rate | Answers naming your brand ÷ all answers to the question set | How often a shopper asking these questions hears your name |
| Share of mentions | Mentions of your brand ÷ mentions of all brands in the same answers | Your weight against competitors, in the way share of voice works |
A worked example: 24 runs of one question set, 14 answers name you, so the presence rate is 14 ÷ 24 = 58%. If those 24 answers name brands 60 times in total and 14 of the mentions are yours, the share of mentions is 14 ÷ 60 = 23%. The two figures answer different questions. We report presence as the headline because a merchant asks “do shoppers hear our name”, and share of mentions beside it for the competitor view.
How many prompt runs do you need?
More than most screenshots suggest, because answers change from run to run. The table gives the 95% confidence interval around a measured rate of 50%, computed with the exact Clopper-Pearson method and rounded:
| Runs | 95% interval around 50% |
|---|---|
| 24 | 29% to 71% |
| 50 | 35% to 65% |
| 100 | 40% to 60% |
| 200 | 43% to 57% |
| 400 | 45% to 55% |
Our July 2026 audit of a chocolatier near Geneva measured a 58% presence rate in Google AI Overviews over 24 prompts, with an interval from 36% to 77% (chocolatier use case). An interval that wide is the honest result of 24 runs. To see a change of ten points, plan on a few hundred runs per assistant and period.
How is share of model different from share of voice?
Share of voice counts a brand’s presence in media or ads it can buy or earn. Share of model counts presence in answers generated by a model, which a brand cannot buy, since ChatGPT states its product results are not ads, and which change when the model or its retrieval changes. The second definition above borrows the share of voice formula and applies it to AI answers.
What does a good report contain?
Five readings, each with its interval where it is a rate:
- Presence, the share of answers that name you.
- Position, where you appear when you are named.
- Competitors, who is named instead of you and how often.
- Sources, the pages the assistant cited, which is where effort pays.
- Accuracy, whether the price and facts stated about you are correct, since a wrong price stated by an agent remains the merchant’s problem.
For a brand that also sells through retailers, the competitors reading shows whether the assistant names the brand’s own shop or a reseller, which is the question behind agentic commerce for brands selling direct.
What can share of model not tell you?
It does not measure revenue, and it does not explain on its own why a number moved. Assistants change models and retrieval without notice, so a drop can come from the engine rather than from your pages. The 2026 survey by Olivier Martinez, covering 45 studies, found no technique with a stable effect across platforms (arXiv 2607.14035), which is why we report per assistant and read changes over several months.
Autobasket Discover runs this measurement as a retainer, and the Buyer Journey Check gives a one-off baseline. For what moves the rate, read how to get your products recommended by ChatGPT, and for the term in one line, the glossary.