When AI Says It Is Not Sure: AI Uncertainty Expression, Trust, and Purchase Hesitation in Online Health Product Recommendation

Abstract

AI recommendation assistants increasingly provide quasi-advisory support in online health-product shopping, yet little is known about how their expressions of uncertainty shape consumer decisions. Drawing on the stimulus–organism–response framework, signaling theory, and computer-mediated communication/Computers Are Social Actors logic, this study conceptualizes AI uncertainty expression (AIUE) as a message-level cue of epistemic certainty. A 2 (AIUE: high vs. low) × 2 (health risk severity: high vs. low) between-subjects experiment with 599 online consumers showed that AIUE reduced competence trust and benevolence trust and increased purchase hesitation. Trait indecisiveness moderated both trust-formation paths: the negative effects of AIUE were stronger among consumers low in indecisiveness. Conditional indirect-effect analyses showed that competence trust transmitted the effect at both low and high indecisiveness, whereas benevolence trust did so only at low indecisiveness. The focal model was estimated in a design containing both mild and more serious HRS contexts. Because the manipulation also shifted perceived recommendation strength, helpfulness, and informativeness, the downstream effects should be interpreted as responses to naturally expressed uncertainty rather than to an isolated certainty cue. These findings extend research on AI-mediated communication by showing how uncertainty wording shapes distinct trust judgments and purchase hesitation and by identifying trait indecisiveness as a boundary condition.