Consumer Acceptance of AI-Powered Personalized Recommendations in E-Commerce: The Mediating Role of Trust and Perceived Value

Abstract

Artificial Intelligence (AI) has revolutionized the eCommerce sector with its rapid implementation of recommendation systems, allowing for highly personalized shopping experiences. While many consumers feel comfortable with AI-generated recommendations, others are less so, likely based on a blend of trust, value, and privacy considerations. This study examines how perceived personalization may affect consumers' acceptance of the AI recommendation, as well as the mediating role of consumer trust and perceived value, and the negative effect of consumers' privacy concern on their trust in the AI. The Technology Acceptance Model (TAM), Trust Theory and Value Co-creation Theory were used in order to develop a conceptual framework and test it empirically.

The methodology adopted for this research is quantitative research design and the tool used is a structured questionnaire administered to 400 online shoppers. The analysis was performed using Partial Least Squares Structural Equation Modelling (PLS-SEM) in SmartPLS 4. The results indicate that perceived personalization has a significant effect on consumer trust (β = 0.43, p < .001) and perceived value (β = 0.38, p < .001). Consumer trust (β = 0.31, p < .001) and perceived value (β = 0.36, p < .001) positively affect consumers' acceptance of AI recommendations. Moreover, consumer trust has a significant impact on behavioral purchase intention (β = 0.29, p < .001). Non-transparent data practices negatively affect trust formation (β = −0.22, p < .001), underscoring the need for transparent data practices. Both consumer trust and perceived value contribute significantly to the relationship between perceived personalization and recommendation acceptance, as shown by the mediation analysis.

The study makes a significant contribution to the literature by integrating technology acceptance, trust, and value-based approaches into a single, empirically validated AI recommendation acceptance model — a dual-mediation configuration that recent 2025 studies have called for but not yet tested in a single design (Frimpong-Manso et al., 2025). The results have real-world applications for e-commerce platforms aiming to improve user engagement through personalized suggestions while preserving trust and protecting user privacy. As generative-AI-based personalization becomes more prevalent, the study's emphasis on transparent data practices and value-centric design offers a timely, practically actionable path to sustained consumer acceptance of AI-driven recommendation systems.