What Makes AI Name You? Determinants of Local Business Recommendation Across Five Generative Models

This study uses an automated, API-only pipeline to measure which observable business and web-entity signals predict whether a generative AI model recommends a specific local service business by name. Five models (GPT-4o, Claude Sonnet 4.5, Gemini 2.5 Pro, Sonar Pro, Grok 4) are queried across 18 U.S. cities (stratified by population), 5 service categories, and 8 intent-diverse query variants, with 3 repeated runs per condition within a 48–72 hour window. A parallel business audit pipeline collects structured signals (reviews, ratings, web presence, schema markup, authority proxies) for candidate businesses. Primary outcomes include mention probability, recommendation probability, rank position, citation rate, and cross-repeat stability. Analysis uses hierarchical mixed-effects models with pre-specified robustness checks. Total planned prompts: 10,800. All methods, prompts, and scoring rubrics will be published. The study is designed for reproducibility and peer-reviewed submission.