Share of model
Share of model is the proportion of AI answers in a category that mention or cite your brand or products, measured by running a fixed set of buyer prompts across assistants on a schedule and counting appearances. It is an industry-coined metric with no standard definition and no first-party reporting behind it, so two vendors' numbers for the same brand are rarely comparable.
What it is trying to measure
It is the AI-answer analogue of share of search or share of shelf: of all the answers a buyer could get in your category, what fraction include you. The appeal is obvious, because AI answers name a short list rather than returning ten links, and being off that list is closer to being absent than to ranking eleventh.
The mechanics are simple. Define a prompt set, define the assistants and models, run them on a cadence, and record for each response whether the brand was mentioned, whether a product was named, and whether a URL was cited. Divide.
Why the number is softer than it looks
Four sources of variance, none of them small.
Prompt set selection. The score is entirely determined by the questions asked. A set weighted toward your strongest categories produces a flattering number that means nothing.
Non-determinism. The same prompt returns different answers and different sources on repeat runs. Single-run measurements are samples, and small prompt sets have wide confidence intervals nobody reports.
Personalization and geography. Answers vary by account history, region, and whether web search was invoked.
Model churn. Providers ship new model versions continuously. A step change in your score may be a model update rather than anything you did.
None of this makes the metric useless. It makes it a directional instrument that requires a frozen prompt set, repeated runs, and stated methodology to be worth anything.
Making it useful in B2B
The consumer version asks "best running shoes for flat feet" and counts brands. That framing transfers badly to distribution, where the buyer's question is a specification question and the winning answer is a part number.
A more useful panel for a distributor is built from real demand: top quote-request phrasings, the application questions inside support tickets, part-number cross-reference lookups, and the top unbranded category queries from Search Console. Score three things separately rather than collapsing them — brand mentioned, specific SKU named, and your URL cited — because they move independently and the third is the one tied to attribute coverage.
Segment by category too. A single blended percentage across a 400,000-SKU catalog hides exactly the information you need, which is which categories are answerable from your data and which are not.
What to say about it in a board deck
State the method next to the number. Prompt count, assistants tested, run frequency, date the set was frozen. A share-of-model figure without that context is not comparable to last quarter's, let alone to a competitor's.
And be clear about what it does not measure. It does not measure revenue, does not measure clicks, and does not capture answers inside a buyer's own procurement assistant, which for many distributors is where agent-mediated purchasing will actually happen.
Frequently asked questions
What is share of model?
The percentage of AI answers across a defined prompt set that mention or cite your brand or products. It is measured by re-running the same prompts across assistants on a schedule and counting appearances. There is no standard definition, so methodology has to be stated alongside the number.
Is share of model the same as share of search?
It is the same idea applied to a different surface. Share of search counts query volume or visible results in a category; share of model counts appearances in generated answers. Share of search draws on query data, while share of model is estimated from sampling, which makes it considerably noisier.
How many prompts do you need for a reliable score?
More than most panels use, and the honest answer depends on how variable the responses are. Because the same prompt returns different sources on repeat runs, both prompt count and repeat runs per prompt matter. Freeze the set, repeat it, and report trends rather than single readings.
Do AI providers report share of model?
No. None of the major assistants publish a visibility console equivalent to Google Search Console, so every share-of-model figure comes from third-party sampling. Vendor numbers for the same brand differ because their prompt sets and counting rules differ.