Study Shows Over 80% of ChatGPT Shopping Decisions Originate from Product Cards

On October 8, 2026, local time, ReFiBuy commissioned Clickstream Solutions to release the findings of an AI shopping behavior study. The study did not rely solely on questionnaires or traffic statistics; instead, it tracked the complete behavioral trajectories of 40 U.S. participants as they completed 224 shopping tasks across 6 product categories in ChatGPT and Google AI Mode, covering the entire process of comparison, exclusion, and selection. Simultaneously, verbal protocols were collected to reconstruct decision-making logic, with the research focusing on the entry-point value of product cards in AI shopping scenarios. The statistical results revealed that 84% of final product selections in the ChatGPT context came from product cards embedded in the conversation, which directly display product images, prices, and key specifications. Items not included in the product card display pool were largely missed during the critical decision-making and comparison phase. Moreover, 75% of shopping tasks began with user interaction with product cards, and front-end information such as images, titles, and prices significantly influenced early decision-making. Positional factors had a pronounced impact on conversion. Calculations by Eric Van Buskirk, founder of Clickstream Solutions, showed that when the AI assistant displayed two or more product cards, the selection rate for the first card was 43.4%, compared to just 29% in random-selection scenarios, with the second card's selection rate even falling below random expectations. Virtually all positional advantages tilted toward the first slot. Overall, 43% of product selections landed on the first displayed product card. In the context of offer selection, 76% of choices favored the first displayed offer card, and the price, stock status, and shipping information presented on these cards directly dictated users' subsequent actions. In supporting materials accompanying the study, Scot Wingo, co-founder and CEO of ReFiBuy, noted that product cards within ChatGPT function similarly to Amazon's Buy Box, serving as the core vehicle for securing consumer purchase decisions. Ensuring accurate product-to-card matching and securing higher placements in offer card lists are essential actions for merchants adapting to the ChatGPT ecosystem. For brands focusing on answer engine optimization (AEO) and generative engine optimization (GEO), gaining exposure is merely the starting point; the ultimate conversion competition unfolds at the product card and offer card levels. ReFiBuy defines optimization at this stage as agentic commerce optimization (ACO), and based on the research findings, has formulated eight actionable recommendations for brands and retailers, covering strategies such as pushing products into the product card display pool, prioritizing first-position placement, and maintaining consistency in price and specification details across product cards, offer cards, and product detail pages. The full report also includes session transcripts from shopping interactions, elements influencing user decisions, user reactions to sponsored placements, and examples of price mismatches on product cards discovered during decision-making. This article was first published on the official website of Ebrun.