BENCHMARK 2026March 15, 202612 min read

Share of Model (SoM) Benchmark: Empirical Visibility Across 5 Foundation Models

A longitudinal audit of 500,000 synthetic enterprise queries evaluating brand presence across ChatGPT, Gemini, Claude, Perplexity, and Copilot.

Giovanni Cocco
Giovanni CoccoApplied Research Lead
BENCHMARK TELEMETRY & EMPIRICAL RIGOR
Queries Audited500,000+
Foundation Models5
Confidence Interval99.4%
UI Divergence Rate41.8%
EXECUTIVE BRIEFING

Executive Summary

As consumer and enterprise decision-making migrates from search engines to Generative AI systems, traditional search volume metrics (Search Share) have ceased to reflect actual brand reach.

Over 50% of B2B and B2C procurement journeys now begin with an LLM dialog interface. In this paper, Cocco Applied Research presents the findings of our 500,000-query longitudinal study measuring Share of Model (SoM) across the five dominant foundation platforms: OpenAI ChatGPT (GPT-5 & Canvas), Google Gemini 3, Anthropic Claude 5 (Artifacts & Sonnet), Perplexity AI, and Microsoft Copilot.


1. The Share of Model (SoM) Formulation

Traditional Search Share measures ranking in an inverted index. In contrast, Share of Model evaluates the probabilistic likelihood $P(Brand \mid Context, Intent)$ of a brand being retrieved, synthesized, and favorably recommended by an autoregressive neural network.

$$\text{SoM}{\text{Brand}} = \frac{\sum{i=1}^{N} w_i \cdot \mathbb{I}(\text{Brand Recommended}i)}{\sum{i=1}^{N} w_i \cdot \sum_{k} \mathbb{I}(\text{Competitor}_k \text{ Recommended}_i)}$$


2. Key Empirical Findings

2.1 The Native UI vs. Developer API Divergence

Our multi-platform telemetry identified a systematic 41.8% divergence between responses generated via public developer APIs and those observed inside live consumer web and workspace interfaces.

Foundation PlatformProduction UI ModelAPI Base ModelSystem Prompt InjectionUI Feature Layer
ChatGPTGPT-5 Enterprisegpt-4o / gpt-5-previewHigh (Search Routing)Canvas, Memory
ClaudeClaude 5 Sonnetclaude-sonnet-latestHigh (System Directives)Artifacts Engine
GeminiGemini 3 Ultragemini-1.5-proExtreme (Workspace Data)Live Google Graph
PerplexitySonar Pro Enterprisesonar-mediumReal-time Web SynthesisMulti-index Search

Critical Insight: Brands measuring their AI visibility solely through API scraping are making strategic capital allocations based on inaccurate synthetic approximations.

Tags:#Share of Model#GEO#Foundation Models#Empirical Benchmarking
RELATED RESEARCH

Continue Reading

Need to apply these findings across your enterprise?

Schedule a technical session to discuss the methodology of this research and evaluate its direct application to your systems and enterprise architecture.