Signal Study Finds AI Search Models Name Individual Professionals in Just Over a Quarter of Responses, With Sharp Language and Model Gaps

Summary

A study measured how often AI models "name" individual professionals — rather than firms — in grounded search-style responses, issuing 2,400 API calls across four models (GPT-5.6 Sol, Gemini 3.6 Flash, Perplexity Sonar Pro, and Grok 4.5) in four European markets and five query languages during a two-hour window on July 24, 2026. Overall, models named a specific individual in 25.8% of responses, but the rate varied sharply by category: real estate agents were named in 35.4% of responses and car dealership staff in 32.9%, compared with only 9.1% for insurance. The four models diverged nearly four-fold, from Grok at 38.0% to Gemini at 9.3%. Naming was predicted by citation type — responses that named an individual cited that person's own site or a category portal more often — but not by overall citation volume. On matched translation pairs, English-language prompts named an individual in 36.7% of responses versus 15.6% for the identical question asked in the local language. A roster of 939 professionals built from public LinkedIn search matched only 128 of 27,293 name-shaped mentions (0.47%), leading the authors to conclude that roster-based measurement captures only a small, unrepresentative slice of actual AI-driven individual visibility.