When help feels creepy: A MOA model of AI smart-retail adoption via perceived intrusiveness and the moderating role of trust in AI

AI is reshaping in-store and omnichannel retail, yet adoption hinges on whether assistance feels helpful or “too close.†This study develops and tests a perceived-intrusiveness (PI)–centered MOA framework for AI-enabled smart retail. PI is theorized as the proximal appraisal that links motivation, opportunity, and ability cues to attitude and, in turn, adoption intention. Specifically, perceived usefulness and personalization quality (motivation), AI transparency and user control (opportunity), and algorithmic literacy and privacy self-efficacy (ability) are modeled as antecedents of PI. Using quota-based data from Taiwanese shoppers who recently used AI retail services (n = 861), we estimated PLS-SEM models with 5000-sample bootstrapping. Personalization quality and user control are the strongest intrusiveness-reducing antecedents; ability variables exert smaller but significant negative effects; usefulness and transparency are weaker. Trust in AI moderates the first-stage links: it strengthens the intrusiveness-reducing effects of personalization quality and user control, does not amplify usefulness, and transparency shows a compensatory pattern (helping chiefly when trust is low). All six antecedents exhibit significant indirect effects operating through PI to shape attitudes and, sequentially, intention. An exploratory check distinguishes personalization intensity from quality and finds a positive incremental link to PI net of quality. The findings clarify designable levers and boundary conditions in MOA and yield actionable guidance: prioritize quality over intensity, provide granular in-flow controls, and use purpose-linked transparency especially for low-trust segments.