Heuristic or systematic? understanding consumer information processing of ChatGPT recommendations

Abstract

Consumers increasingly rely on generative AI tools such as ChatGPT for product and service recommendations, yet little attention has been paid to how they process and evaluate ChatGPT outputs. To fill this gap, we examine whether users rely on heuristic versus systematic processing modes when interpreting ChatGPT outputs, especially when ChatGPT generates recommendations. Across three scenario-based experiments in product purchase and travel decision contexts, we find a robust first-item preference: consumers disproportionately choose the first-listed recommendation even when item order is randomized, numerical ranking cues are removed, or the top-listed option includes an error. Decision-time evidence is consistent with low-effort, position-based processing rather than careful comparison of option attributes. We also identify regulatory focus as a moderator of this processing tendency. Compared to promotion-focused consumers, prevention-focused consumers are less likely to rely on the first-listed option, distribute their choices more evenly across alternatives, and spend more time deliberating. These findings extend dual-process perspectives to AI-mediated recommendation environments, showing that positional reliance emerges even when no explicit ranking rationale exists, which is a different phenomenon from position effects in search engine contexts where ranked results carry meaningful quality signals. The study also establishes regulatory focus as a theoretically grounded moderator that reveals the processing mechanism underlying positional reliance, and offers implications for the design of e-commerce recommendation interfaces that encourage more careful consumer evaluation.