Human-like, Animal-like, or Object-like? The Impact of LLM-Based Virtual Doctor Avatar Design on User Emotion, Physiology, and Experience

Virtual agents powered by large language models are increasingly deployed in digital mental health services, yet the influence of avatar appearance on users’ emotional, cognitive, and physiological responses remains insufficiently understood. This study examined how three avatar designs—animal-like, human-like, and object-like—shape affective experience, user evaluation, autonomic activity, and attentional allocation during virtual doctor interactions. Forty-two participants completed a within-subjects experiment involving self-reported affect ratings, multidimensional user-experience assessments, heart rate variability (HRV) measures, and eye-tracking indicators. Avatar type did not significantly affect changes in positive or negative affect. However, physiological data revealed clear divergences. The animal-like avatar elicited the strongest parasympathetic activation, reflected by significant increases in RMSSD and HF power, whereas the object-like avatar produced a sympathetic-dominant response. Across six user-experience dimensions, the animal-like avatar consistently received the highest evaluations. Eye-tracking results showed faster first fixation and longer face-directed fixation duration for the animal-like avatar, indicating stronger social attention. The human-like avatar demonstrated slightly delayed initial fixation, consistent with subtle yet nonsignificant uncanny-valley tendencies. These findings underscore the critical role of avatar visual design in shaping emotional safety, engagement, and social processing in virtual mental health interactions.