AI-DRIVEN HYPER-PERSONALIZATION IN DIGITAL MARKETING AND ITS IMPACT ON CONSUMER TRUST, PURCHASE DECISIONS, AND LONG-TERM CUSTOMER LOYALTY IN E-COMMERCE PLATFORMS: A MULTI-THEORY AND EXPLAINABLE AI PERSPECTIVE

Abstract

Artificial intelligence (AI)–driven hyper-personalization has become the dominant paradigm through which e-commerce platforms curate product recommendations, pricing, and content at the level of the individual consumer. Yet the same algorithmic precision that increases relevance also intensifies perceptions of surveillance, manipulation, and opacity, producing what the literature has termed the “personalization-privacy paradox.” This paper addresses a persistent gap in the marketing and information systems literature: the absence of an integrated, theory-driven account of how algorithmic transparency — operationalized through Explainable AI (XAI) — conditions the relationship between hyper-personalization and downstream relational outcomes, namely consumer trust, purchase decisions, and long-term customer loyalty. Drawing jointly on the Technology Acceptance Model, Privacy Calculus Theory, Trust Transfer Theory, and Uses and Gratifications Theory, we develop and propose an empirical test of a moderated-mediation model in which perceived personalization value and perceived algorithmic intrusiveness jointly shape consumer trust, with XAI-enabled explanation quality moderating the strength and even the sign of these paths. We specify a mixed-methods design combining a large-scale cross-sectional survey (structural equation modeling), a between-subjects online experiment manipulating explanation transparency (SHAP-based versus black-box recommendation interfaces), and computational text analysis of consumer reviews. We report illustrative structural estimates to demonstrate the analytic approach and interpret the theoretical implications of the proposed model, showing that explanation transparency is expected to function as a boundary condition that converts personalization from a trust-eroding to a trust-enhancing mechanism once a threshold of perceived intrusiveness is reached. The paper contributes a falsifiable, multi-theory model of algorithmic trust formation, a validated multi-item explanation-quality construct, and actionable design principles for responsible hyper-personalization. Limitations and directions for longitudinal and cross-cultural replication are discussed.