Measuring AI Share of Model (SoM): How to Quantify Brand Mentions Across LLMs at Scale
AI Share of Model (SoM) is a performance metric that quantifies how frequently a brand is mentioned, cited, or recommended by Large Language Models (LLMs) relative to its competitors. It is measured by running standardized prompt sets across multiple AI engines and calculating the percentage of responses that include the brand, often weighted by the sentiment and position of the mention.
Measuring AI Share of Model (SoM): How to Quantify Brand Mentions Across LLMs at Scale
As generative AI replaces traditional search for a growing percentage of users, the industry is shifting from Share of Voice (SoV) to Share of Model (SoM). While SoV measures visibility in search engine results pages (SERPs), SoM measures a brand's presence within the latent space and output of LLMs.
What is AI Share of Model (SoM)?
AI Share of Model is the quantitative measure of a brand's visibility within the responses generated by AI answer engines. Unlike traditional SEO, where success is measured by rankings and click-through rates, SoM focuses on "mention frequency" and "recommendation probability."
If a user asks an AI, "What are the best CRM tools for small businesses?" and the AI lists five tools, a brand appearing in that list has captured a portion of the SoM for that specific intent. When aggregated across thousands of prompts, this reveals the brand's overall dominance within the model's knowledge base.
How to Quantify Brand Mentions at Scale
Measuring SoM manually is impossible due to the non-deterministic nature of LLMs—meaning the same prompt can produce different results each time. Scaling this measurement requires a systematic, programmatic approach.
1. Establishing a Prompt Library
To get an accurate reading, you must create a comprehensive library of prompts that mirror actual user behavior. These should be categorized by: * Direct Queries: "What is [Brand Name]?" * Comparative Queries: "[Brand A] vs. [Brand B] for [Use Case]." * Category Recommendations: "What are the top-rated tools for [Industry]?" * Problem-Solution Queries: "How do I solve [Problem] using [Category]?"
2. Programmatic Execution via API
To quantify mentions at scale, marketers use API integrations (such as OpenAI's GPT-4 or Anthropic's Claude) to run these prompt libraries repeatedly. By running each prompt 10 to 50 times, you can establish a statistical baseline of how often the model chooses to mention your brand.
3. Analysis of Citation and Sentiment
A mention alone is not enough. To calculate a true SoM score, the data must be filtered by: * Presence: Did the brand appear? (Binary: Yes/No) * Position: Was the brand the first recommendation or an afterthought? * Sentiment: Was the mention positive, neutral, or critical? * Citation: Did the AI provide a link to the brand's website?
The SoM Formula
While there is no single industry-standard equation, a foundational approach to calculating SoM for a specific category is:
SoM = (Total Brand Mentions / Total Competitor Mentions in Category) × 100
To refine this, advanced practitioners use a weighted score where a "top-of-list" recommendation carries more weight than a mention in a concluding paragraph.
Why Brands Fail to Capture Share of Model
If your SoM is low despite high traditional SEO rankings, it is likely due to a gap in your Generative Engine Optimization (GEO) strategy. LLMs do not just look at keywords; they look for consensus, authority, and structured relationships between entities.
Common reasons for low SoM include: * Lack of Third-Party Validation: LLMs rely on "consensus." If your brand is not mentioned in reputable industry lists, forums, or news articles, the AI views you as a lower-authority entity. * Poorly Structured Data: AI engines struggle to parse ambiguous data. Implementing AI-friendly structured data helps the model definitively link your brand to specific solutions. * Low Citation Frequency: If your brand is not cited across the diverse datasets used to train the model, it will not be recommended. This is why understanding how to get your brand cited by ChatGPT is critical for increasing SoM.
SoM vs. Traditional Share of Voice (SoV)
| Feature | Share of Voice (SoV) | Share of Model (SoM) |
|---|---|---|
| Primary Metric | Impressions, Rankings, Clicks | Mentions, Recommendations, Citations |
| Discovery Path | Search Engine Results Page (SERP) | Conversational AI Response |
| Driver | Backlinks, Keywords, Page Speed | Entity Authority, Consensus, GEO |
| User Intent | Navigational or Informational | Decision-making or Synthesis |
Implementing an AI Presence Audit
To move the needle on your SoM, you must first understand where you stand. An AI presence audit involves testing your brand across multiple engines—including ChatGPT, Claude, and Perplexity—to identify "blind spots" where competitors are being recommended over you.
AI Presence provides the strategic framework and tools necessary to bridge this gap. By analyzing the delta between your current SoM and your target SoM, brands can deploy targeted LLM optimization tactics to increase their visibility.
Key Takeaways
- SoM is the new SoV: It measures the probability of a brand being recommended by an LLM.
- Scale requires APIs: Manual prompting is insufficient; programmatic testing across a prompt library is the only way to quantify SoM accurately.
- Consensus is King: LLMs recommend brands that appear consistently across high-authority, third-party sources.
- Weighting Matters: Not all mentions are equal; the position and sentiment of the mention determine the actual value of the SoM.
- GEO is the Lever: To increase SoM, brands must shift from traditional SEO to Generative Engine Optimization.