How AI-Driven Recommendation Accuracy and Novelty Shape Online Purchase Intention: The Mediating Roles of Perceived Value and Trust

Abstract

Artificial intelligence recommendations must do more than reproduce a consumer’s established preferences: they must also facilitate worthwhile discovery without undermining confidence in the platform. Drawing on the stimulus–organism–response framework, this study examines recommendation accuracy and recommendation novelty as two distinct system cues, perceived value and trust as parallel cognitive mechanisms, and online purchase intention as the behavioral response. The model was tested with 493 complete survey responses and 27 reflective indicators using partial least squares structural equation modeling in SmartPLS 4. Recommendation accuracy and novelty were positively related to both perceived value and trust. Perceived value, in turn, had a positive relationship with purchase intention. Trust also had a positive but comparatively weak relationship with purchase intention; this path met the conventional percentile-bootstrap criterion, although its bias-corrected confidence interval included zero. The specific indirect effects through perceived value were supported for both accuracy and novelty, whereas the corresponding trust-based indirect effects were not supported under conservative inference. The model explained 7.5% of perceived value, 3.4% of trust, and 4.0% of purchase intention, indicating that recommendation quality is a meaningful but limited part of the wider purchase decision. The findings distinguish relevance from discovery and show that the value pathway is more dependable than the trust pathway in translating AI recommendation quality into purchase intention.