AI Commerce Exposure
A longitudinal exmxc framework for measuring which retailers enter the consideration set created by consumer AI systems, how prominently they rank, and how recommendation visibility relates to machine access, product legibility, agentic-commerce integration, and transaction architecture.
Definition
AI Commerce Exposure measures whether an economic entity enters the consideration set created by consumer AI systems, and how prominently it is ranked once surfaced.
The framework separates seven layers that should not be collapsed into a single score: external AI access, Entity Clarity, product legibility, agentic-commerce integration, recommendation presence, transaction accessibility, and owned AI-commerce capability.
Why it matters
AI shopping is moving toward a mainstream discovery layer. NielsenIQ reported on September 24, 2026 that 51% of U.S. consumers had used at least one AI-powered shopping tool in the prior month. As AI increasingly filters the consideration set before a consumer reaches a retailer, recommendation visibility becomes an economic variable distinct from conventional search rank or transaction share.
Core measures
- Recommendation Share: eligible valid runs in which a retailer appears as a positive ranked recommendation divided by all category-eligible valid runs.
- First-Position Share: eligible valid runs in which a retailer ranks first divided by all category-eligible valid runs.
- Rank Distribution: the distribution of observed positions among recommendations.
- Platform Divergence: variation in recommendation presence across consumer AI systems.
Research architecture
Recommendation presence is an outcome variable. It is never inferred from robots.txt, Entity Clarity, structured product data, partnerships, or retailer scale. Those variables are measured separately and tested against recommendation outcomes.
The first benchmark population is the frozen NRF 2026 Top 30 U.S. retailer set. Wave 1 uses 13 frozen shopping prompts across ChatGPT, Perplexity, Gemini, and Claude, with three replicates per prompt-model pair.
Longitudinal design
Wave 1 is the baseline. exmxc will repeat the frozen recommendation experiment every two weeks and append each wave to a longitudinal series. Protocol changes require a new version; historical waves are never rewritten.
The objective is to observe whether retailer recommendation share changes around shifts in machine access, product legibility, agentic-commerce integrations, model behavior, and transaction architecture.
Strategic interpretation
The framework distinguishes external AI distribution openness from owned AI-commerce capability. This makes it possible to study a new strategic separation: discovery control versus transaction capture.
Research boundary
AI Commerce Exposure is not a universal retailer-quality score and does not prove causality. Results are conditional on the tested prompts, consumer interfaces, model states, accounts, and collection windows.