Title: When Relevance Becomes Work: Cognitive Load and Trust in the AI Personalization-Engagement Relationship
Authors: Tran Nguyen Gia Thinh, Nguyen Viet Trung Nhan, Nguyen Manh Ha
Volume: 10
Issue: 9
Pages: 239-248
Publication Date: 2026/09/28
Abstract:
Personalization is the most widely deployed AI consumer application, built on the logic that closer matching increases engagement. However, evidence reveals a downturn at high intensities. This paper proposes that personalization imposes a processing cost alongside relevance, and that trust moderates whether this cost drives withdrawal, positioning trust on the cost path rather than as a benefit-side mediator. Across an experiment manipulating personalization intensity (N = 360) and a two-wave observational panel (N = 421), engagement exhibited an inverted U-shape. Crucially, engagement fell at high intensity while perceived usefulness continued rising, locating the downturn on the cost side rather than a failure of relevance. This processing cost is highly asymmetric, driven overwhelmingly by presentation-specific extraneous load, while a transparency control moved no load indicators. Furthermore, cognitive load translated into disengagement significantly less among higher-trust respondents (+.112 experimentally; +.188 in the panel), attenuating from ?.436 at low trust to ?.061 at high trust. The position of the engagement optimum did not move with trust. Trust therefore operates as a defensive shield against processing costs rather than extending users' tolerance for dense personalization. The paper notes the load path accounts for a minority of the downturn as its principal limitation.