Share of Model is a brand’s share of recorded brand mentions within a defined prompt sample. Report the denominator, engines and period with the percentage, and keep the definition consistent over time. It measures presence in sampled answers, not market share, product quality or the probability of a purchase.
What is Share of Model, exactly?
Share of Model is a ratio. The numerator is the number of times your brand is mentioned across a fixed set of category-defining prompts. The denominator is the total number of brand mentions in those same answers. Multiply by 100 and you have a percentage that compresses everything an AI model thinks about your category into one number.
Two things distinguish it from earlier visibility metrics. First, the audience is not human – it is the model. The "respondent" is ChatGPT, Perplexity, or Google AI Overviews, sampled the way a research firm samples consumers. Second, the answer is multimodal: a defensible Share of Model number is computed across at least three AI platforms, because no single model represents the AI search market. BrightEdge found platforms disagreed on brand recommendations for 61.9% of identical queries. A SoM tracked on one model is a vanity number.
Where Share of Model came from
The term was coined by Jack Smyth, Chief Solutions Officer at Jellyfish. Jellyfish launched the Share of Model platform on December 4, 2024, with Danone and Chivas Brothers as beta participants. The original framing in Adweek was simple: "Measuring how each model perceives your brand, compares it to competitors and why it suggests your products to customers will become an essential responsibility for every marketing team."
Eighteen months later, Boston Consulting Group adopted the same vocabulary in its 2026 brief on agentic scenarios for marketers, listing Share of Model alongside position, citations, sentiment, and referral traffic as the KPIs brands need to monitor when AI agents start making purchase decisions. Hallam, Symphonic Digital, and a long tail of GEO publishers have since picked it up. The metric is moving from agency thought-leadership into board decks – which is why senior marketing leaders should learn it before someone else does.
How Share of Model differs from Share of Voice and Share of Search
The three metrics measure different surfaces of the same problem. Share of Voice came from advertising: how much paid presence does my brand have versus competitors. Share of Search measures organic intent: what fraction of category searches mention my brand. Share of Model measures generated answers: what fraction of the brands an AI recommends are mine. Each is downstream of a different consumer behavior – paying, searching, asking.
| Metric | What it measures | Source data | What it predicts |
|---|---|---|---|
| Share of Voice | Paid media presence vs competitors | Ad spend, impressions, mentions | Brand awareness lift |
| Share of Search | Branded organic search volume vs competitors | Google Trends, search query data | Future market share |
| Share of Model | AI-generated brand mentions vs competitors | LLM responses to category prompts | AI-driven discovery share |
The metrics do not replace each other. They stack. A brand winning on Share of Voice but losing on Share of Model is buying awareness that AI assistants no longer route to it.
How to calculate Share of Model
The arithmetic is straightforward; the discipline is not. The formula is:
Share of Model = (mentions of your brand ÷ total category brand mentions) × 100
The work is in the inputs. To produce a number that holds up under scrutiny, you need:
- A frozen prompt set of 50–500 category-defining queries (size depends on category breadth)
- A frozen competitor set so the denominator is stable across runs
- Multiple AI platforms, sampled the same way every time (ChatGPT, Google AI Overviews, Perplexity at minimum; add Claude and Gemini when defensible)
- Repeated runs per prompt – at least three samples per prompt per platform – because LLM outputs vary
- Temperature set to 0 where the API allows, so randomness is bounded
- A counting rule that explicitly handles parent-brand vs sub-brand mentions, ranking position, and lists vs prose
Without the discipline, two reports of "Share of Model" can disagree by 30 points on the same brand because they used different prompt sets and different sampling rules. The metric is only useful if your team writes the rules down and re-uses them.
How many prompts and how often you should sample
The right prompt set is small enough to be defensible, large enough to be representative, and stable enough to be comparable. Most practitioner teams converge on a similar shape: 50 to 200 prompts split across category, comparison, problem-solution, and recommendation intents. Symphonic Digital recommends 250–500 high-intent queries for a full SoM index; Hallam suggests 5–10 informational prompts is enough for a starter dashboard. Both are right, for different audiences.
Sampling cadence matters more than most teams expect. AI answers shift week to week – only about 30% of brands stay visible across consecutive monitoring runs. Quarterly snapshots will lag reality by a quarter. Practical cadence: sample weekly for the operational team, roll up monthly for the executive view, and produce a quarterly snapshot for the board. If you only have time for one cadence, pick weekly – you can downsample later.
If you want to know when AI changes its answer about your brand, start with a free brand check — it takes a minute.
What to do when AI models disagree on the answer
Disagreement between models is not noise. It is signal. BrightEdge's 61.9% disagreement number means that on most category prompts, ChatGPT and Google AI Overviews recommend different brands. A single weighted Share of Model number will smooth that disagreement and hide the most actionable insight in your data: which platform is winning for your brand and which is winning for your competitors.
Track Share of Model per platform first. Roll up to a weighted total only after you have decided how much each platform matters for your business. A B2B SaaS brand's customers may live primarily in ChatGPT and Perplexity, where the answer set is narrower (averaging 2.37 brands per query in BrightEdge's analysis). A consumer brand's customers may live in Google AI Overviews, which mentions 6.02 brands per query. Same Share of Model number, two different competitive realities.
How to report Share of Model to your CEO
Executives do not want a research deck. They want one number, one trend line, and one decision. A useful executive Share of Model report has four slides: today's number, the four-quarter trend, the gap to the leader and the runner-up, and a short list of what would move the number next quarter.
The framing that works: "Share of Model is our share of brand recommendations across AI assistants. It is to AI search what Share of Voice was to advertising." Then anchor the number against a benchmark the executive already knows – your Share of Voice in your category – so they can compare apples to apples. If your Share of Voice is 18% and your Share of Model is 4%, you have an investment misallocation, not a metric problem. We covered the rest of this conversation in our piece on how to report AI visibility to your CEO.
Where Share of Model breaks down
Share of Model summarizes your presence but leaves out useful detail. It does not tell you whether the mention was a recommendation or a warning (being named and being the pick are different metrics). It does not capture whether your brand was first or fifth in the list, or how crowded that list is in the first place (a typical answer names only about five brands). It does not flag when an AI model has confused your brand with a competitor of the same name. And it can be inflated by mentions in adjacent categories that do not convert.
The fix is not to abandon the metric. It is to pair it with three companion measures: rank position within an answer, sentiment of each mention, and citation source – the third-party domain the model leans on for the answer. Parse tracks AI visibility across ChatGPT and Google AI Mode, and reports those companion measures alongside the headline Share of Model number. If you only put one number on the board slide, that is fine. If you only track one number, your operating team will fly blind.
How Share of Model fits into the rest of your AI visibility stack
Share of Model is the headline number. The work that moves it sits underneath: prompt selection, content optimization, citation source coverage, entity disambiguation. A brand that improves its Share of Model usually does it by getting cited by more of the third-party domains that AI models trust – review sites, comparison pages, Reddit threads, established publications. We covered the source mechanics in which domains AI models cite most and the prompt-set engineering in how to build an AI visibility prompt set.
The category has produced a lot of names – AI Share of Voice, GEO Score, Generative Visibility Index, LLM Mention Rate – most of which are repackagings of the same arithmetic. Share of Model has won the naming battle because it is short, it parallels Share of Voice, and BCG put it in published work. The cleanest mental model: Share of Model is the brand-level KPI; AI visibility is the broader discipline; citation share, mention rank, and sentiment are the diagnostic dimensions. Use Share of Model when reporting up; use the underlying dimensions when deciding what to fix.
What is the difference between Share of Model and Share of Voice?
Share of Voice measures paid advertising presence relative to competitors. Share of Model measures the percentage of brand mentions inside AI-generated answers that belong to your brand. Share of Voice predicts brand awareness lift; Share of Model predicts AI-driven discovery share. They are complementary, not interchangeable. A brand can be loud in advertising and invisible in AI answers, or vice versa.
How many prompts do I need to track Share of Model?
Fifty to two hundred is the practical range for most categories. Smaller sets (5–20) work for a starter dashboard but are too noisy for executive reporting. Larger sets (500+) are defensible for an indexed product but expensive to run. The selection process matters more than the size: the prompts must represent how real buyers ask, not what your team wishes they would ask.
Which AI platforms should I include in Share of Model?
At minimum: ChatGPT, Google AI Overviews, and Perplexity. Add Claude and Gemini when your audience overlaps with their user base. Track each platform separately first, then roll up to a weighted total. BrightEdge's data showing 61.9% platform disagreement on brand recommendations means a single-platform Share of Model overstates or understates your real position by a wide margin.
How often should Share of Model be measured?
Weekly for the operations team, monthly for the executive view, quarterly for the board. AI answers shift faster than quarterly reporting can capture – only about 30% of brands stay visible across consecutive monitoring runs. If you can only afford one cadence, pick weekly; you can always downsample for the slide deck.
Is Share of Model the same as AI Share of Voice?
In practice, most vendors use the terms interchangeably. The original distinction matters: Share of Voice is the parent concept (presence relative to competitors); Share of Model specifies the surface (LLMs) and the unit (brand mentions in generated answers). Pick whichever name your organization will commit to and stick with it for at least four quarters.
Can I calculate Share of Model manually without a tool?
Yes – for 5–10 prompts on 2–3 platforms, a spreadsheet works. The math is mention counting. The constraint is repeatability: rerunning manually next quarter introduces enough variance to invalidate the trend line. Automation matters once you cross 25 prompts or you start reporting Share of Model to executives who will ask quarter-over-quarter questions.
Share of Model is now the number senior marketing leaders are expected to know for their own brand. The underlying mechanics – prompt selection, multi-platform sampling, mention counting – are the same mechanics behind every defensible AI visibility program. If you have done the foundation work, the headline number falls out of it. If you want a Share of Model number this quarter without building the pipeline yourself, that is what we built Parse for.
