What Is Share of Model? The Metric That Tells You Whether AI Knows You Exist
Key takeaways
|
Somewhere right now, a VP at a company you'd love to work with is typing a question into ChatGPT. Something like, what are the best options for a mid-market team that needs to fix its content pipeline? About four seconds later, they have a shortlist. Three to five names, a line about each, maybe a caveat or two. They'll copy it into a doc, forward it to a colleague and start their evaluation from there.
You won't know that conversation happened. There's no impression in Search Console, no click in Analytics, no keyword you could have bid on. You were either on that list or you weren't, and if you weren't, the deal may already be over before your sales team knew it existed.
That's the problem share of model was invented to describe. It's the most important number most B2B marketing teams still aren't tracking, and the teams that start now will have a year of trend data by the time their competitors start asking what it is.
What is share of model?
Share of model is the percentage of AI-generated answers, across a defined set of category prompts and AI assistants such as ChatGPT, Claude, Gemini and Perplexity, that mention or recommend a specific brand. It's the AI-search counterpart to share of voice, measuring whether AI engines include your brand when buyers ask about your category. |
Rented attention versus earned attention
Share of model borrows its logic from one of the oldest metrics in marketing. Share of voice compares your presence in a category with your competitors' presence, usually through advertising spend, search visibility or media mentions. Mark Ritson, writing about AI agents in 2026, described share of model as exactly that, the AI-era equivalent of share of voice, as The Drum reported.
But there's a difference that matters more than the similarity. Much of share of voice can be bought. You can outspend a competitor on paid search, sponsor more events, run more ads. Share of model mostly can't be bought. You can't bid your way into ChatGPT's recommendation. You get there because enough credible sources across the web describe you as belonging there, and because your own content answers the questions buyers actually ask.
The way I put it to clients is simple. Share of voice is rented, share of model is earned. That makes it harder to move quickly. It also makes it far more durable once you've moved it, and much harder for a better-funded competitor to take away.
It's also worth separating share of model from citations, because people blur them constantly. A citation is your page being linked as a source. A mention is your brand being named in the answer. You can have one without the other, and the mention is what the buyer remembers. Brand mentions in AI answers goes deeper on why that distinction matters.
Why one prompt tells you almost nothing
The most common mistake I see is treating a single AI answer as evidence. Someone on the leadership team asks ChatGPT about the category, doesn't see the company named, and forwards the screenshot with a one-word message. Someone in marketing runs the same question an hour later, gets a different answer that does include the company, and screenshots that. Both are real. Neither means much.
AI answers are probabilistic. When SparkToro and Gumshoe had around 600 volunteers run the same brand-recommendation prompts, the same list of brands came back less than once in a hundred runs, and the same list in the same order less than once in a thousand. That's the phenomenon covered in AI non-determinism, and it's why share of model has to be measured as a rate across many prompts, many runs and several assistants, never as a single answer.
Here's the reassuring part of that same research. Even though the lists and the order bounced around, the brands that genuinely owned a category kept showing up. Variation sits on top of a stable core. Share of model is really a measure of whether you're part of that core, and whether you're moving toward the middle of it or drifting out to the edges.
How to calculate it without fooling yourself
The arithmetic is trivial. The discipline is not. You start by writing a fixed set of buyer questions, somewhere between thirty and a hundred, in the language buyers actually use rather than the language your product team prefers. A good set mixes early discovery questions, shortlist questions and head-to-head comparison questions. AI search prompts covers how to find that real buyer language.
Then you run every prompt several times across several assistants, at minimum ChatGPT, Perplexity, Google's AI features and one of Claude or Gemini, and you log every brand named in every response. Share of model is simply the number of responses that name you divided by the total number of responses. If you run fifty prompts three times on four assistants, you have six hundred responses. Appear in 132 of them and your share of model is 22%. If your biggest competitor appears in 246, theirs is 41%, and now you know the size of the gap and exactly where it lives.
The discipline is in keeping that set stable. If you change your prompts every month, you lose the trend, and the trend is the whole point. Add new prompts as the market evolves, but keep a core that never changes, and run it on a schedule.
Score it on four dimensions, not one
A single percentage hides too much. Imagine a brand that's named in nearly every answer, but always with the same line, expensive and hard to implement. That brand has an impressive share of model and a serious problem. So I score four things, which I think of as the four Ps.
Presence is the headline number, and it asks whether you're named at all. Position asks whether you're named early, in the opening summary or as the first option, or buried at the bottom of a long list. Portrayal asks whether the description is accurate and favorable, or carries an outdated or damaging caveat. Proof asks whether your own pages are the cited source behind the mention, or whether a review site or competitor is telling your story for you.
Presence tells you where you stand. The other three tell you what to fix. Weak Position usually means you're known but not strongly associated with the core use case. Weak Portrayal usually traces back to old reviews, dated comparisons or plain hallucinated facts. Weak Proof means your content isn't answering the question cleanly enough to be the source. AI sentiment analysis covers how to score Portrayal in practice.
What a good share of model looks like
I'd be skeptical of anyone selling a universal benchmark right now. Categories differ enormously in how concentrated they are. Similarweb's 2026 AI Brand Visibility Index found category leaders ranging from 15.89% in finance, where Chase led, to 54.38% in consumer electronics, where Apple led. Most B2B categories are more fragmented than either, and the leader often has far less.
So I judge it three ways, against direct competitors, against your own trend over time, and against your market share. That last comparison is the interesting one. Les Binet and Peter Field's work for the IPA found that brands whose share of voice exceeds their share of market tend to grow, and brands whose share of voice falls below it tend to shrink. My working hypothesis, and I want to be clear it's a hypothesis, is that share of model behaves the same way. If AI names you more often than your market share would predict, you're building future demand. If it names you less, a competitor is quietly building it instead. Share of voice in AI search explores that idea further.
What actually moves the number
Because share of model reflects what the web says about you, most of the levers sit off your own site. Ahrefs' study of 75,000 brands found that branded web mentions correlated with AI Overview visibility at 0.664, compared with 0.218 for backlinks. That's a striking gap, and it points to where effort pays, in review profiles, trade press, analyst notes, podcasts, community discussions and the comparison articles buyers read when they're building a shortlist.
On your own site, two things matter most. The first is consistency, meaning the same name, description and category everywhere, backed by Organization schema and sameAs links, which is the territory of entity authority for B2B SaaS. The second is distinctiveness, meaning named frameworks, original data and genuine points of view that give a model a reason to associate you with a topic, rather than content that restates what everyone else says. That's the heart of LLM SEO.
Expect two speeds. Answers built from live search, the process described in search grounding, can shift within weeks of new coverage. Answers drawn from a model's training data move on the model's release cycle, which can take many months. Both matter. The first is where you'll see early wins. The second is where lasting position is built.
Taking it to leadership
The best share of model reports I've seen fit on one slide. This quarter versus last, with your top three competitors alongside. The four Ps underneath. The handful of prompts where you gained or lost the most, and what you're doing about each. That's it.
The framing matters as much as the numbers. Pipeline tells leadership what already happened. Share of model tells them whether you'll be on the shortlist for the deals that haven't started yet. In a category where buyers increasingly start their research by asking an AI, that's not a vanity metric. It's an early warning system. Content marketing KPIs shows where it sits on the wider dashboard, and AI search visibility tracking covers the tools that can automate the work.
Share of model in the Engine Optimization Matrix™
Share of model is the scoreboard for the Citation column of the Engine Optimization Matrix™, especially in the GEO and LLMO rows, which ask whether your brand is named in generative engine outputs, and whether ChatGPT, Claude or Perplexity name you specifically when prompted. Every other cell in those rows exists, in one way or another, to move this number.
If you only track one AI visibility metric this year, make it this one. For the surrounding ideas, see AI visibility, what is zero-click search and why ChatGPT recommends your competitors.
Share of model is the scoreboard. Mastering AI Visibility is the playbook behind it, with AI SEO, AEO, GEO and LLMO in one Demand Doctrine™ edition, built on the Engine Optimization Matrix™. Out November 18.
Frequently asked questions
What is share of model?
Share of model is the percentage of AI-generated answers in a category that mention or recommend your brand, measured across a defined set of prompts and AI assistants.
How is share of model different from share of voice?
Share of voice measures your share of attention across media and search, much of which can be bought. Share of model measures whether AI assistants name you, which mostly has to be earned.
How do you calculate share of model?
Run a fixed set of buyer prompts several times across several AI assistants, count the responses that name your brand and divide by total responses.
How many prompts do you need to measure share of model?
Thirty to 100 prompts, each run at least three times per assistant, gives a reasonable picture. Fewer prompts produce mostly noise.
What is a good share of model?
There's no universal benchmark. Similarweb's 2026 index found category leaders ranging from 15.89% in finance to 54.38% in consumer electronics. Compare against competitors and your own trend.
Why does share of model change between runs?
AI answers are probabilistic. SparkToro and Gumshoe found the same brand list repeated less than once in 100 runs, which is why many runs are needed.
How do you improve share of model?
Earn mentions on trusted sites, keep entity data consistent, publish named and citable ideas and structure content so AI search can retrieve it.
Which tools track share of model?
Tools such as Semrush's AI Visibility Toolkit, Profound and Peec AI track brand mentions across AI assistants. A spreadsheet and a fixed prompt set work too.
About the author. Harold Bell is the founder of MQL Magnet and has spent about 16 years in enterprise technology marketing, working with brands like Cisco, Rubrik and Nutanix in different capacities. He's a member of the Forbes Communications Council, the creator of the Engine Optimization Matrix™ and the author of Partner Over Product. |
