Prompt to Purchase: ChatGPT Referrals, Basket Structure, and Category Concentration in E-Commerce

This paper examines whether purchases referred by ChatGPT differ systematically from those generated through traditional e-commerce acquisition channels. Using transaction-level data from seven online shopping portals between January 2025 and February 2026, covering more than 11 million baskets and approximately 117 million purchased items, we analyze how ChatGPT referrals relate to basket size, basket revenue, within-basket category diversity, and the concentration of demand across product categories. We combine basket-level OLS models with store fixed effects and a hierarchical multinomial logistic framework that identifies categories disproportionately associated with specific marketing channels. Although ChatGPT accounts for less than 0.5% of transactions in every store, ChatGPT-referred purchases exhibit a distinct structure. They contain fewer unique items, show lower within-basket category diversity, and generate slightly higher revenue per transaction than direct-access purchases. At the category level, ChatGPT displays an intermediate concentration pattern that differs from both broad-spectrum channels such as direct access and organic search and narrowly targeted channels such as affiliate traffic. Specifically, ChatGPT referrals are characterized by a concentrated head of strongly overrepresented categories alongside a meaningful long tail of additional categories. These results suggest that generative AI already functions as a distinct form of shopping intermediation, shaping consumers’ consideration sets before they reach merchant websites and reallocating demand across product space in ways that differ from established digital channels.

Working Paper