Share of model is the AI search version of share of voice. The idea is sound. The measurement is young, and anyone telling you otherwise is selling something.
Share of model is how often an AI assistant names your brand when someone asks about your category. If you run a London accountancy firm and ChatGPT is asked to recommend accountants in London, your share of model is the proportion of those answers that mention you. It is the AI search equivalent of share of voice.
The idea is sound and the metric is young. Nobody has a reliable way of measuring it yet, and that is worth saying plainly before anyone quotes you a number.
Why the metric exists at all
Traditional SEO measures positions on a results page you can see and screenshot. AI assistants don’t work that way. There is no ranked list, no stable position and often, no click. A buyer can ask an assistant for three suppliers, get three names, then contact one without ever visiting a search engine. If your brand isn’t in that answer, you were never in the running, and no rank tracker will tell you.
Share of model is the attempt to put a number on that. It answers one question: when the model speaks for our category, how often does it say our name?
How people measure it
The method is broadly the same everywhere. Write a list of prompts a real buyer might type, usually 30 to 100 of them. Run each prompt across the assistants that matter to you, several times each. Count how often your brand appears and how often each competitor does. Divide your mentions by total brand mentions.
Tools including Ahrefs Brand Radar, Profound and Peec do this at scale. The output is a percentage plus a trend line that looks reassuringly like a rank-tracking chart.
Why the number is rougher than it looks
- The models are not deterministic. Ask the same question ten times and you can get ten different brand lists. Small sample sizes produce noise that looks like movement.
- Prompt choice decides the answer. Change “best accountants in London” to “affordable accountants in London” and the whole set of names changes. Whoever picks the prompts is largely picking the result.
- Personalisation. Memory, location, account history and browsing context all shift answers. A clean test environment is not what your buyers are using.
- No benchmark. Nobody knows what a good share of model is for a given category, because there is no shared methodology and no published baseline.
Using it without fooling yourself
Treat it as a directional signal, not a KPI you report to the board. Fix your prompt set and your sampling method, then leave both alone so the trend means something. Watch the competitor list more than your own percentage: knowing which four brands the model reaches for is more useful than knowing you are at 11%.
The work that shifts it is mostly familiar. Get mentioned on the sources models draw from, keep your entity data consistent, and publish content that answers the questions directly. We build this into AI search work for clients rather than selling it as a separate product, because the levers overlap heavily with ordinary SEO and digital PR.
Common questions
Is share of model the same as share of voice?
Same idea, different surface. Share of voice measures your presence across a media or search landscape. Share of model measures your presence inside AI-generated answers. The maths is similar; the data collection is much harder.
What is a good share of model?
No established benchmark exists, and be wary of anyone quoting one. It depends entirely on how many credible competitors exist in your category and which prompts you sampled. The useful comparison is your own number over time, measured the same way each month.
Which AI assistants should I measure?
Start with whichever your customers actually use. For most UK B2B businesses that means ChatGPT first, then Google AI Overviews, then Gemini and Perplexity. Measuring five platforms badly is worse than measuring two carefully.
Can I influence share of model directly?
Only indirectly. Models draw on training data, on live retrieval from the web, and on the sources they cite. Earning mentions in the publications and listicles they pull from is the closest thing to a direct lever, and it takes months.
See our glossary entry for share of model for the short definition.
