The Right Amount of “Human Touch”: The Curvilinear Effect of Embodied AI Agent Anthropomorphism on Consumer Usage Intention under Social Intrusiveness

Anthropomorphic design in embodied AI agents often evokes ambivalent consumer responses, yet it remains a key strategy for enhancing users’ intention to use such agents. Based on three video-based scenario experiments with 438 participants, this study examines how anthropomorphism shapes consumers’ usage intentions. The results confirm an inverted-U-shaped relationship between anthropomorphism of embodied AI agents and usage intention. Further analyses reveal that likability and eeriness jointly mediate this curvilinear relationship: anthropomorphism exerts an inverted U-shaped effect on likability, which in turn positively influences usage intention, whereas anthropomorphism exerts a U-shaped effect on eeriness, which negatively influences usage intention. In addition, social intrusiveness functions as a boundary condition by (1) strengthening the curvilinear relationships between anthropomorphism and both likability and eeriness, and (2) moderating the indirect effects of likability and eeriness. These findings provide empirical support for applying social information processing theory to human–AI interaction and extend uncanny valley theory from emotional responses to usage intention. The study also offers practical implications for deploying embodied AI agents in customer service contexts.