Share of Model (SoM)

Table of Contents

  1. Definition
    1. How It Relates to Marketing
  2. How to Calculate Share of Model
  3. How to Utilize Share of Model
  4. Comparison to Similar Approaches
  5. Best Practices
  6. Future Trends
  7. Frequently Asked Questions
  8. Related Terms
  9. Sources

Definition

Share of Model (SoM) is a marketing metric that measures how often, how prominently, and how favorably a brand appears in answers generated by large language models, relative to its competitors. It’s the AI-era counterpart to share of voice. Where share of voice tracked how much of a category’s advertising or organic search presence a brand owned, Share of Model tracks how much of an LLM’s output a brand owns when someone asks ChatGPT, Gemini, Claude, or Perplexity a category question.

Put concretely: if a shopper asks an assistant “what are the best CRM platforms for small businesses?”, Share of Model tells you whether your product shows up in the answer, where it lands in the response, how it’s characterized, and how that stacks against rivals. The metric has a couple of related senses worth separating. The original framing, introduced by Jellyfish executives Jack Smyth and Tom Roach in 2024, looked at how brands are represented inside a model’s knowledge — what associations and attributes a model carries. The measurement-and-GEO framing, which gained traction through 2025 and 2026, focuses on live responses: the share of brand mentions a company earns across a sample of category prompts. Researchers at INSEAD formalized the term in mid-2025, defining it as a measure of how often, prominently, and favorably brands appear in AI-generated responses.

Back to Top

How It Relates to Marketing

The reason marketers care is that LLMs have become gatekeepers to discovery. Roughly half of consumers already use AI-powered search, and McKinsey has projected AI search will influence around $750 billion in revenue by 2028. ChatGPT reached close to 883 million monthly users, Gemini’s share climbed from about 5% to 21% in a year, and Google’s AI Overviews now appear in nearly half of all searches. When that much of the buying journey runs through synthesized answers, the old scoreboard stops working. There’s no “position one” to win, and click-based traffic becomes a lagging signal of something that already happened inside the model.

Share of Model also behaves differently from a ranking, which trips people up. A keyword rank is static — you’re number one or you aren’t. SoM is probabilistic. A brand might surface in 80% of responses to “best organic skincare” and 20% to a near-identical prompt, and raising that probability is the whole point of generative engine optimization. It’s also fragmented across models in ways that defy a single “AI visibility” number. The INSEAD study found the detergent brand Ariel held close to 24% Share of Model on Meta’s Llama but under 1% on Gemini. A brand isn’t visible in AI generally — it’s visible in specific models, shaped by what each was trained and grounded on.

Back to Top

How to Calculate Share of Model

There’s no single ratified method yet, but the dominant approach borrows from election polling. You assemble a representative set of prompts a real customer might ask in your category, run them repeatedly across the major models, and count how the brands come up. Repetition matters because the same prompt returns different answers across sessions, so a single query tells you almost nothing.

The core ratio is straightforward:

Share of Model (%) = (mentions of your brand across category responses ÷ total brand mentions in those responses) × 100

Most practitioners break the measurement into three levers rather than collapsing it into one figure:

  • Mention frequency, sometimes called the recall or inclusion rate — the percentage of relevant prompts where your brand appears by name at all. This is usually the baseline you establish first.
  • Prominence — where and how the brand shows up, since being the first recommendation isn’t the same as a passing mention buried in a list.
  • Favorability — the sentiment and attributes attached to the brand, because appearing often but described poorly isn’t a win.

A worked example: run 100 category prompts across four models, 10 times each, for 4,000 responses. If your brand is named in 1,200 of them and all competitor brands together account for 6,000 total mentions, your raw inclusion rate is 30% and your Share of Model is 20%. Tracking both numbers over time, per model, is more useful than chasing one blended score. Tools like Profound, Semrush, Adobe LLM Optimizer, and Jellyfish’s Share of Model platform automate the prompt runs and tallying, which is the only practical way to do this at scale — manual spot-checks are too small a sample and too biased to trust.

Back to Top

How to Utilize Share of Model

The first use is a baseline and competitive benchmark. Knowing you appear in 20% of category answers while a rival sits at 55% reframes the problem in a way leadership understands, and it’s specific enough to act on. From there, SoM becomes the target metric for GEO work: you change content, earn mentions, fix structured data, then re-measure to see whether the probability moved.

It works as a diagnostic too. Because the metric is per-model, a brand that’s strong in ChatGPT but invisible in Gemini learns exactly where to focus. Sentiment and attribute tracking surface a different problem — when a model consistently describes a product with the wrong features or an outdated price, that’s a grounding issue to correct at the source. Several platforms now pipe SoM into Looker Studio or GA dashboards alongside traffic and conversion data, so teams can watch whether visibility shifts track with downstream revenue.

Common drivers of the metric give the work direction. Recent analyses suggest third-party brand mentions correlate with AI visibility roughly three times more strongly than traditional backlinks, which pushes effort toward earned media and digital PR. User-generated content — reviews, ratings, photos — acts as a freshness and verification layer that models lean on to confirm a brand’s current status. And specific, data-rich content tends to get pulled into answers more readily than vague marketing copy.

Back to Top

Comparison to Similar Approaches

MetricWhat it measuresWhereNature
Share of ModelBrand mentions, prominence, and favorability in AI answersLLMs and AI assistantsProbabilistic; varies by model and session
Share of VoiceBrand’s slice of total advertising or media presencePaid and owned mediaSpend- or impression-based
Share of SearchBrand’s slice of category search queriesSearch enginesDemand-side, query volume
Keyword RankingPosition of a page for a querySearch engine resultsStatic position (1st, 2nd, etc.)

The closest relative is share of voice, and the leap is from a deterministic count of impressions to a probabilistic measure of whether a model chooses to mention you. Keyword ranking is the sharpest contrast: a rank is a fixed slot you either hold or don’t, while Share of Model is a likelihood that shifts with phrasing, model, and time. Some practitioners use the label “Share of Model Voice” (SoMV) for the same idea, and “AI share of voice” shows up as a near-synonym in vendor tools.

Back to Top

Best Practices

  • Sample broadly and repeat. Run many category prompts across every model that matters to your audience, multiple times each, and treat the output as directional trends rather than exact figures.
  • Measure per model, not as a single blended score, since visibility in one model says little about another.
  • Track frequency, prominence, and favorability separately so a high mention count doesn’t hide poor positioning or wrong information.
  • Invest in earned third-party mentions and credible UGC, which move the metric more than backlinks alone.
  • Feed models accurate, specific, current information at the source to reduce hallucinated attributes and stale claims.
  • Tie SoM to downstream metrics like AI referral traffic and conversion, so the score connects to business outcomes instead of sitting as a vanity number.
  • Re-measure on a regular cadence, because model updates and training refreshes can shift results without warning.
Back to Top

Several forces are pushing Share of Model from a niche idea toward a standard line on the dashboard. Search volume itself is expected to shift — Semrush has predicted AI search traffic will overtake traditional search by 2028, and some analyses see organic search volume falling around 25% as answers replace clicks. As that happens, a metric built for AI answers becomes harder to ignore.

A few open problems will shape how the metric matures. Measurement standardization is the big one: methods, sampling, and even definitions still vary between vendors, and no accuracy benchmark exists yet. Temporal decay is another, since a model’s knowledge can reflect a brand’s past market position rather than its current one, especially with older training cutoffs. Hallucination remains a persistent source of noise, both for measurement and for brands that get described incorrectly. Looking further out, the metric will likely extend from text answers into agentic commerce, where the question isn’t only whether a model mentions a brand but whether a shopping agent selects it — a shift that ties Share of Model directly to the visibility frameworks emerging around agent-mediated buying.

Back to Top

Frequently Asked Questions

1. What is Share of Model in simple terms? It’s the share of AI-generated answers in your category where your brand appears, weighted by how prominently and favorably it shows up. Think share of voice, but for ChatGPT, Gemini, Claude, and Perplexity instead of ads.

2. Who came up with the term? Jellyfish executives Jack Smyth and Tom Roach introduced the concept in 2024, and researchers at INSEAD formalized the definition in mid-2025.

3. How is Share of Model different from a keyword ranking? A ranking is a fixed position you hold or don’t. Share of Model is probabilistic — your brand might appear in most answers to one prompt and few answers to a similar one, and the figure changes by model and over time.

4. How do you actually measure it? By running a representative set of category prompts across the major models repeatedly and tallying how often each brand is mentioned. Specialized tools automate this, since manual checks sample too little to be reliable.

5. Why does my Share of Model differ between models? Each model was trained and grounded on different data, so visibility is model-specific. A brand can hold a large share in one model and almost none in another, as the INSEAD Ariel example showed.

6. What improves Share of Model? Earned third-party mentions, credible user-generated content, and specific, data-rich information that models can ground against. These tend to matter more than backlinks for AI visibility.

7. Is Share of Model accurate? It’s directional. Answers vary between sessions, methods aren’t standardized, and hallucinations add noise, so it’s best used to spot patterns and track change rather than to fixate on a precise number.

8. Does Share of Model replace SEO? No. It sits alongside SEO as a measure for AI answers. Traditional search still drives substantial traffic, and the two disciplines increasingly inform each other.

Back to Top
Back to Top

Sources

Back to Top

Tags: ,

Was this helpful?

Previous

Product Feed Optimization for AI

Next

Shopping Agent