Share of Model is the share of AI-generated answers that name your brand when people ask about your category, compared with your competitors. It’s the AI-era version of Share of Voice, and you can measure it, as long as you stop treating one ChatGPT screenshot as data.
- Share of Model is your brand’s mention rate in AI answers, measured over fixed prompts, platforms and dates.
- You can’t buy it: only about 12% of AI-cited URLs also appear in Google’s top 10.
- Benchmarks range from 6.8% to about 31% median, so track your baseline weekly for four to six weeks.
- Twenty to thirty prompts and a spreadsheet get you started; paid trackers begin around $29 a month.
What Is Share of Model, and How Do You Calculate It?
Share of Model measures how often, how prominently and how favorably your brand shows up in answers from large language models, compared with your competitors. In practice, you run a fixed set of category prompts across several AI platforms, count the answers that mention you, and compare that rate with the rates of the brands you compete against.
The term isn’t a tool vendor’s invention. Jellyfish’s Jack Smyth and Tom Roach coined it in 2024, and in July 2025 INSEAD researchers formalized the concept together with Jellyfish co-authors. Their version tracks three dimensions: mention rate, the gap between human and AI awareness of a brand, and brand sentiment.
The formula: Share of Model (%) = answers that mention your brand ÷ all answers checked × 100. A second version divides by all brand mentions in your category instead, which gives you your slice of the conversation rather than your appearance rate. Both versions are in circulation, so always state which denominator you used.
Here’s a worked example: 20 prompts across four engines give you 80 answers, and if your brand appears in 14 of them, your Share of Model is 17.5%.
Share of Model vs. Share of Voice: Why It Can’t Be Bought
Share of Voice tracks how much of the market conversation you own, and you can push it up with ad spend. Share of Model tracks how often AI models choose to name you, and no media budget buys that directly.
| Share of Model | Share of Voice | AI visibility | |
|---|---|---|---|
| What it measures | How often LLM answers name your brand vs. competitors | Your share of ads, mentions or impressions in your category | Umbrella term for mentions, citations and links in AI answers |
| Denominator | All answers from a fixed prompt set (or all brand mentions in it) | Total category media or conversation volume | Varies by tool |
| Can you buy it? | No | Partly, through paid media | No |
Two other terms you’ll hear in vendor decks are related but distinct. Share of search counts branded search volume, so it tracks demand. Brand mentions counts raw references across the web, without a fixed prompt set or a competitor comparison.
The reason you can’t buy Share of Model comes down to where models get their information. Only about 2.9% of AI citations point to a brand’s own domain, while 75.2% come from third-party sites. A correlation study of 75,000 brands found that branded web mentions correlate with AI Overview visibility at 0.664, while backlinks reach only 0.218, roughly a 3:1 gap. Its authors stress that correlation isn’t causation, but the direction is clear: models trust what others say about you. That’s the core shift behind generative engine optimization.
Building Your Prompt Set: 20-30 Prompts That Actually Track Something
Start with 20 to 30 prompts that real buyers would type, run them weekly on the same platforms, and give it at least a month before you draw conclusions. A Share of Model number only counts as a measurement when it states its prompt set, platform list and time window. Without those three, it’s anecdotal.
Split your set roughly into thirds: generic category questions (such as “best CRM software for small teams”), comparison questions (such as “HubSpot vs. Pipedrive vs. Zoho for a 10-person sales team”), and problem or use-case questions (such as “how do I stop leads going cold after a demo”).
To pick the prompts, look at the questions that come up in sales calls, support tickets and your Search Console queries. Leave out branded prompts like “what is [your brand]”: you’ll almost always show up there, and they inflate the number without telling you anything about consideration.
For platforms, cover ChatGPT, Perplexity, Gemini and Google AI Overviews if your buyers use all four. Track them separately even if you report one combined figure later. INSEAD’s analysis of Italian detergent brands found Ariel at nearly 24% on Meta’s Llama but under 1% on Gemini, and a combined average would hide exactly that gap. Weekly is the right cadence for a small team: often enough to spot a trend, and rare enough that it doesn’t eat your Mondays.
Why repeat at all? SparkToro’s own conclusion from its research is that a visibility percentage across dozens or hundreds of repeated prompts is meaningful, and a single sample isn’t.
What to Log Each Run (and How to Avoid False Noise)
Log seven fields every time, in one spreadsheet row per prompt per platform: prompt (exact wording, copied rather than retyped), platform and model (if visible), date, brand mentioned (yes or no), position or prominence (first named, in a list, footnote only), cited URL (if any), and sentiment note (positive, neutral, negative, plus a few words on how you’re described).
The controls matter as much as the columns. SparkToro’s study of 2,961 prompt runs across ChatGPT, Claude and Google AI found that the odds of getting the same brand list twice for an identical prompt are under 1 in 100. Getting the same order twice is closer to 1 in 1,000. So you remove every source of variance you can control.
Run prompts logged out or in an incognito window, so chat memory and personalization don’t tilt the result toward brands you’ve researched before. Keep the wording identical every week. If you sell into DACH, run the full set in both German and English, since many B2B buyers prompt in English, and answers can differ a lot between languages. Log the language as its own field so you never mix the two in one number.
Which Tools Track Share of Model, and What Do They Cost?
Your zero-cost option is the spreadsheet from the last two sections, and for your first few rounds it’s often the smarter choice: you learn which prompts matter before you pay to track them. Once manual runs take more time than the insight is worth, these are the named options at current list prices:
| Tool | Tier | Entry price per month | Good fit if… |
|---|---|---|---|
| HubSpot AEO Grader, Mangools AI Search Grader | Free one-off check | Free | You want a first snapshot without tracking over time |
| Otterly.ai | Entry | $29 | You want the cheapest ongoing tracker |
| Mentionable (Growth) | Entry | €79 | You prefer EUR billing and a simple setup |
| Peec AI (Starter) | Entry | €85–95, 50 prompts | Your prompt set fits within 50 prompts |
| Semrush AI Visibility Toolkit | Mid | $99 per domain | Your team already works in Semrush |
| Ahrefs Brand Radar | Mid | $199 | Your team already works in Ahrefs |
| AthenaHQ | Free to mid | Free (300 credits) to $295 (3,600 credits) | You want to test before committing |
| Profound (Growth) | Agency | $399 | You track several brands or clients |
Above this range, enterprise suites and several tracking tools only quote on request. If you’re testing whether the metric matters to your business at all, pick a monthly plan without an annual commitment.
Your First Numbers: What’s Normal, What’s Noise?
No single “good” Share of Model exists, and anyone who quotes one without a methodology is guessing. Published benchmarks put the cross-industry median at 6.8% in one dataset, an average of 12–15% in another, and a median of around 31% in a third. That’s not a contradiction: each tool uses different prompts, platforms and definitions of a mention. Compare yourself with your own baseline and your two or three closest competitors.
A study of 17,691 prompt runs across 20 brands shows why averages mislead smaller brands: aggregate mention rates reach 26.8% on Gemini, 21.1% on Perplexity, and 16.9% on ChatGPT, but median per brand is only 1.1% on Gemini, 0.7% on Perplexity, and 4.2% on ChatGPT. The top 20% of brands took about two-thirds of all visibility, and 36% of brand-and-engine pairs scored exactly 0%.
Worth knowing: If you’re a small or mid-size brand and your first run shows low single digits, you’re likely sitting near the real median. A 0% on one platform is common, and it’s the starting point for a baseline.
For week-to-week judgment, two findings help. In SparkToro’s data, the top brands in a category still appeared in 55–77% of answers regardless of prompt wording, so the consideration set is fairly stable even when the order shuffles. Separate tracking data shows 96.8% of citation positions stay unchanged week to week, and 87% of the moves that do happen are declines.
A single dip is mostly noise. Sustained direction over three or four weeks constitutes a trend. Trust your data after four to six weekly runs, and if leadership asks what the number means for the business, pair it with the ways to measure AI visibility beyond traffic.
The Levers That Actually Move Share of Model
The strongest lever is third-party citations on sites models trust: Reddit, industry forums, review platforms, YouTube and trade press. A follow-up analysis from December 2025 extended the same correlation research to ChatGPT and Google AI Mode and found YouTube mentions correlating even more strongly, at around 0.737. Technical SEO factors and sheer page count showed little connection.
Reddit is the most-cited domain across nearly every major AI engine, and one citation analysis puts it at up to one in five citations on Perplexity. Domains with a strong Reddit presence see roughly four times higher AI citation rates. In practice, show up where your buyers already ask questions: answer threads honestly, get your product into category roundups, and pitch data or opinions to trade publications.
On your own site, make content easy to quote. Add comparison tables and FAQ blocks to existing articles before you build new keyword pages, because structured, direct answers are what models pull from.
Your Google rankings won’t carry you either. Only 12% of URLs also rank in Google’s top 10 for the same query, and the most-cited pages tend to have fewer backlinks than pages cited less often. If your reporting rests on rankings, add Share of Model as a separate line rather than assuming one predicts the other.
You can do all of this in-house with the six phases of a GEO framework as your plan. If you’d rather start with a diagnosis, Claire’s Free GEO Audit checks your site against 70+ points that decide whether ChatGPT and other AI systems can find, understand and cite it.
Share of Model Rewards the Long Tail, Not the Loudest
Here’s the fact most Share of Model discussions skip: in one analysis of 200 million prompts over five months, even the most-cited domain on a platform rarely cleared 5% of all citations, and the other 95% was spread across thousands of sources. AI visibility is a long-tail game. For a small or mid-size brand, the realistic goal is a trend line that keeps climbing week after week, built from consistent third-party mentions, rather than category dominance.
Start your spreadsheet baseline this week with 20 to 30 prompts on the platforms your buyers use. If you’d like a faster look at what’s blocking citations today, run Claire’s Free GEO Audit alongside it.
FAQ
Does a brand mention without a link still count?
Yes. A mention counts toward Share of Model because it shapes what the buyer considers, even without a click. Log cited URLs separately, though: they show which sources the model relies on and where your earned-media work should go next.
Does sentiment matter as much as being mentioned?
Usually more, once you’re above the mention threshold. One 2026 analysis found the sentiment framing of an already-cited brand flipped between runs in 45.5% of cases, while presence itself shifted in only 6.8%. That’s why the sentiment column in your log matters.
How do I report Share of Model to management?
Show your weekly trend line next to one or two main competitors, with a footnote stating prompt count, platforms and time window. Don’t promise a target rate or “rank #1 in ChatGPT”, because nobody controls the model’s answers.
When should I not invest in Share of Model yet?
Hold off if your buyers rarely research your category with AI assistants, or if you have no capacity to act on the findings. A metric you can’t influence becomes a vanity number. Run one manual baseline anyway, so you know where you stand.