How Perceived Accuracy Drives Adoption of AI Personalized Recommendations: A Moderated Mediation Model

Artificial intelligence (AI)-powered personalized recommendation systems are reshaping how consumers search, evaluate, and purchase products, yet the psychological mechanisms through which perceived accuracy drives adoption remain underexplored. This study examines how perceived accuracy of AI recommendations influences consumer adoption willingness through perceived benefit and how this process is conditioned by product involvement. Drawing on the Technology Acceptance Model (TAM) and Product Involvement Theory, we develop an accuracy-centred moderated mediation model in which perceived accuracy (PA) leads to perceived benefit (PB), which in turn leads to consumer adoption willingness (AW) or (PA → PB → AW). The study uses survey data from 518 Chinese consumers with experience of using AI-personalized recommendations. The data are analyzed using Partial Least Squares Structural Equation Modelling (PLS-SEM) with multigroup analysis to examine age-based heterogeneity on consumer adoption willingness. The results show that perceived accuracy has a significant direct and indirect effect on adoption willingness, with perceived benefit acting as a partial mediator. Product involvement positively moderates the relationship between perceived accuracy and perceived benefit, and the proposed mechanisms are stable across age groups. The study opens the “black box” linking perceived accuracy to adoption, identifies key boundary conditions, and extends TAM by positioning perceived accuracy as an antecedent of perceived usefulness in AI recommendation contexts.