Xiaoxiao Gong, Yang Zhang
Despite the growing popularity of AI companions, designers have yet to fully consider the critical role of interface modalities in shaping user preferences. We investigate how different interface modalities in AI virtual companion apps (text-based, audio-based, and virtual human-based) influence individual usage intentions. Combining sentiment analysis and the four experiments, we demonstrate that audio-based interfaces elicit the strongest usage intention, outperforming both text-based and virtual human-based interfaces. We also propose that experience perceptions mediate the relationship between interface design and usage intention, highlighting the primacy of psychological factors over visual anthropomorphism—a finding that challenges conventional uncanny valley explanations. Furthermore, attachment avoidance moderates this effect. These findings theoretically advance human-AI virtual companion interaction research by establishing experience perceptions as a mediator and integrating attachment theory. Practically, this study highlights the need for interface designs balancing emotional support with psychological comfort, particularly for users with high attachment avoidance.