Share of Model
Share of Model (SoM) measures how often, and how favorably, generative AI systems like ChatGPT and Gemini mention or recommend your brand versus competitors when answering category-related questions. It has been proposed as a new brand-visibility KPI that succeeds traditional Share of Voice now that AI has become both the source of and the recommender behind the answer.
- Share of Model (SoM) measures how frequently and how favorably generative AI mentions your brand relative to competitors in its answers.
- The concept was introduced by the marketing group Jellyfish with its 'Share of Model™ platform' in late 2024, and the term is a registered trademark.
- Unlike fixed SEO metrics such as keyword rankings, SoM is tracked as a mention rate because LLM responses are probabilistic by nature.
- It is partly qualitative: beyond raw exposure counts, it analyzes the theme, sentiment, and cultural context surrounding each mention.
- As ChatGPT becomes a de facto answer channel, SoM is being discussed as the next-generation visibility KPI that follows Share of Voice.
What Share of Model Is
Share of Model is a metric that captures how often and how positively generative AI systems such as ChatGPT, Gemini, and Llama mention and recommend your brand versus competitors when a user asks a category-related question. The easiest way to grasp it is to think of Share of Voice — which gauged your slice of exposure on the search results page — extended into 'the share you hold within an AI model's answer.' The term was first introduced by the global marketing group Jellyfish with its 'Share of Model™ platform,' unveiled in late 2024, and it is a registered trademark.
The key backdrop is that the path consumers take to discover brands is shifting. Jellyfish frames LLMs as "not merely a tool, but a central part of the customer journey and an audience in their own right." When a user asks an AI for "the most durable luggage brand" and the AI replies by naming a specific brand and summarizing its warranty, that can lead to a purchase without a single search-result click (zero click). In other words, this is an environment where being included in the AI's answer is itself an outcome, every bit as much as ranking in search. That ties directly to the core goal of GEO (generative engine optimization), and Share of Model serves as the unit of measurement that quantifies it.
How It Differs From Share of Voice
Both metrics share the idea of measuring a share, but they differ in what they measure and how they behave. The core distinctions are below.
| Dimension | Share of Voice (traditional) | Share of Model (generative AI) |
|---|---|---|
| What is measured | Share of exposure across search results, ads, and media | Share of brand mentions and recommendations within AI answers |
| Nature of the result | Relatively fixed (e.g., keyword rankings) | Probabilistic — the same question yields different answers each time |
| Unit of measurement | Impressions and share (%) | Mention rate plus sentiment and theme |
| Analytical dimension | Mostly quantitative (whether exposure occurred) | Quantitative (frequency) plus qualitative (sentiment and cultural context) |
| Target channels | Search engines such as Google and Naver | LLMs such as ChatGPT, Gemini, and Llama |
Because LLM responses are probabilistic in particular, a single brand might be mentioned 80% of the time for a "best organic skincare" query — or only 20% of the time. The approach is therefore not a one-off ranking check but repeatedly querying a variety of prompts to track an average mention rate.
Evidence and Examples
According to Jellyfish's announcement, the Share of Model™ platform is "a single-view interface for analyzing how multiple large language models (LLMs) perceive a brand, its products, and its services," with ChatGPT, Google Gemini, and Meta Llama as the models under analysis. As reported by Adweek, the product "measures how frequently AI models mention a brand relative to competitors and analyzes the themes, sentiment, and cultural context," then converts those insights into targeting signals for Google Performance Max (PMax) campaigns. In short, it is built to connect LLM research findings to real-world digital advertising strategy.
The Project Management Institute (PMI) is a frequently cited success case. According to Adweek, this approach led to "a 20% increase in sales, a 45% increase in conversions, and a 156% improvement in return on ad spend (ROAS)" over a roughly 90-day campaign. Danone and Chivas Brothers, among others, took part in the beta, using it to expand keyword strategy and to adjust website text and image assets. Separately, martechview underscored the urgency of managing brand perception inside LLMs, citing that "two-thirds of people aged 18–24 ask AI models for brand recommendations." (Figures are as presented by each source and may vary by campaign and timing.)
Measurement and Activation Checklist
- Define a set of core category questions — gather "best ○○" and "○○ recommendations" prompts that real customers are likely to pose to an AI.
- Run the same prompts repeatedly across major LLMs such as ChatGPT and Gemini, recording how often your brand and competitors are mentioned (measure multiple times to account for probabilistic variation).
- Look beyond whether a mention occurred to the sentiment (positive/negative) and the themes and context it appears alongside.
- Identify the questions where competitors are mentioned more often or more favorably, and use them to diagnose gaps in your content and assets.
- Improve the machine readability of your content so LLMs can cite it easily — apply technical GEO elements together, such as structured data (schema), clear factual statements, and llms.txt.
- Don't treat measurements as one-offs; track them over time to validate the impact of content and campaign changes.