Signal The Beginning of ChatGPT Ads

Summary

Emma Lurie, Ro Encarnación, Sorelle A. Friedler, and Danaé Metaxa published a paper on arXiv (cs.CY) on August 5, 2026, presenting the first empirical study of advertising content rolled out in ChatGPT's user-facing interface. Using a sock puppet audit methodology, the researchers deployed 91 sock puppets in a 3x3 factorial design signaling three racial/ethnic groups (Black, Hispanic, White) and three income terciles (low, medium, high) via geolocation cues. Data collection began in February 2026 and gathered over 3,000 advertisements from 186 unique advertisers in response to 335 prompts covering realistic user queries. The study found accounts begin receiving ads 14 days after account creation, and lower-income accounts, regardless of race, are more likely to receive ads. In this first phase, ads skewed heavily toward consumer goods, directed users to a specific advertiser rather than a particular product, and were clearly separated from the LLM's response text — observations the authors expect to change as ads become more integrated into LLM chat interfaces. The researchers released a public, searchable archive of all collected advertisements.