ReFiBuy announced a comprehensive user behavior study into ChatGPT shopping patterns, conducted with Clickstream Solutions and executed in July and August. The research tracked real consumers navigating ChatGPT's answer engine interface, which comprises three core elements: product cards (the catalog recommendations), written recommendations from the LLM, and offer cards (the sorted list of retailers selling each product) (Retailgentic).
The study revealed that product cards in aggregate draw 75.3% of consumer engagement on the prompt page, with 84% of final product choices originating from a product card (Retailgentic). When multiple cards are displayed, the first position captures 43.4% of selections, position two 17.1%, and position three shows a material bump before declining sharply (Retailgentic). Most striking: 76% of consumers select the first offer on the offer card, with position two dropping to 14% (Retailgentic). For commerce practitioners, this mirrors the "buy box" dynamics that drove Amazon success, establishing ChatGPT offer card position as the new high-stakes real estate. Merchants can improve ranking through offer data accuracy—particularly shipping details, site-wide offers, and return policies—with potential gains of 2–3 positions on the offer card (Retailgentic).