Share of Model (SoM) is an AI visibility metric that measures how often and how prominently a given AI model mentions you across relevant queries — relative to your competitors. It’s the equivalent of Share of Voice, but for AI answers instead of search rankings. And because answers vary from run to run, SoM is a directional trend indicator, not a precise number. This piece covers what SoM is, how to measure it, and how to read it alongside classic search metrics.
Why classic rankings carry less weight
Classic search gives you a ranked list of results. You could measure position, click share, category visibility. In an AI answer you often don’t get a list of results at all — you get a synthesized answer, sometimes with sources attached, in which brands show up with varying prominence.
That changes the question you’re asking. Instead of “where do we rank,” it’s now are we in the answer at all, how prominently, and in what context. That’s exactly what Share of Model tries to capture.
Share of Model vs. related metrics
SoM is easy to confuse with similar-sounding terms. The difference is in what exactly each one measures.
| Aspekt | Metric | What it measures |
|---|---|---|
| Share of Model | Mentions in AI | Your share of the mentions in AI answers across a set of queries |
| Share of Voice | Visibility in a channel | Your share of visibility or mentions in a chosen channel (advertising, organic, media, social) |
| Share of Search | Brand interest | The brand's share of search queries, often used as a directional indicator of interest |
| Mention count | Frequency | How often the model mentions you — with no ratio against the competition |
How to measure Share of Model
You don’t need a paid tool right away. A manual test gives you the basic picture as long as you run it consistently. There’s no settled standard for the calculation yet, so the most important thing is holding the same method over time. Beyond the raw share, you can also score the prominence of the mention separately — whether it was a passing mention, a recommendation, or a cited source.
-
Build a set of real queries
Write down 10 to 30 questions your customers genuinely ask — the right solution for a given situation, not queries on your own name.
-
Run them across models
Put the queries into ChatGPT, Perplexity, and Gemini. Behavior differs between models, so measure each one separately.
-
Record the mentions
For each answer, note whether your brand is mentioned, which competitors appear, and who is listed as a source.
-
Calculate the share of mentions
Divide your brand's mentions by all brand mentions in the answers. Calculate it per model first, then across all of them.
-
Measure regularly and watch the trend
Repeat the same queries over time — monthly, for example. What you care about is the direction of travel, not a one-off number.
Why to take the numbers with a grain of salt
Time to be honest: Share of Model is an estimate, not an exact metric. A few reasons not to trust any single number:
- Non-determinism The same query returns a slightly different answer every time — models don't always generate the same output.
- Personalization and region Results can be shaped by location, language, the mode you used, conversation history, or being signed in; it varies by service and model.
- Models change A model update can shift your SoM overnight without you doing anything at all.
- Small sample = noise A handful of queries isn't enough; even a larger set only gives you a directional estimate.
- A mention isn't a recommendation Distinguish a passing mention, a cited source, and an active recommendation — they aren't the same thing.
What moves Share of Model
When SoM rises or falls, it usually comes down to how much the models have to go on when it comes to recognizing you:
- Citable content — authoritative, well-structured content; see how AI cites sources.
- Third-party mentions — reviews, industry directories, community discussions; for models these are often an important supporting signal alongside your own pages, as brand mentions in AI shows.
- A clear brand entity — a consistent description of the company across your site, so the model connects the mentions to a single subject.
- Presence on sources the models trust — the authorities in your category, not just your own site.
This ties into the zero-click reality: if you aren’t in the answer, you can be less visible to a share of users no matter how well you’re doing in classic search.
The most common measurement mistakes
One measurement = a conclusion
Answers fluctuate. A single snapshot tells you nothing; only repeated measurement and a trend mean anything.
Branded queries
Asking the model about your own name inflates the result. Measure on neutral queries that address a problem, not a brand.
One model standing in for all
Behavior differs between ChatGPT, Perplexity, and Gemini. Averaging across a single model distorts the overall picture.
SoM instead of results
Share of Model is a directional indicator, not a goal in itself. Always pair it with traffic and conversions.
Key takeaways
- Share of Model measures your share of mentions in AI — how prominently the model mentions you compared with the competition.
- Classic rankings alone aren’t enough — in an AI answer, what often decides things isn’t just your search position but whether and how the brand shows up in the answer.
- Measure manually and regularly — a set of queries across models, repeated, with the trend tracked.
- The numbers are directional — between fluctuation and personalization, it’s an estimate, not a precise metric.
- Pair it with results — SoM is a directional indicator, so verify it with the AI segments report in Search Console and traffic from AI.
Want to know where you stand in AI answers compared with your competitors? This site is run by Sniper Design — an AI SEO audit works through your visibility in AI, benchmarks you against competitors on real queries, and shows where you’re missing and how to change it.
For transparency: Share of Model is a metric that’s still taking shape; this description is based on public practice and the tools available as of July 11, 2026. The principles here are directional — specific results vary by category, language, and model, and they shift over time. We describe how the content on this site is produced on the author page.
