“Don't look at me!”: The role of avatars' presentation style and gaze direction in social chatbot design

With improved language ability, social chatbots are capable of forming close relationships with users and empowering various fields. This is particularly true for psychotherapy, where users' self-disclosure is crucial for human-chatbot relationship building and chatbot service provision. Although attention has been given to the role of various anthropomorphic cues in social chatbots, the effects of avatar presentation style and avatar gaze direction on users' self-disclosure have not been well studied. We conducted two experiments to examine how avatar presentation style (profile style vs. background style) can influence users' self-disclosure, and whether the avatar's gaze direction (direct vs. averted) can moderate this relationship. The results reveal that the background-style (vs. profile-style) presentation inhibits users' self-disclosure by decreasing private self-awareness. This relationship is reversed when the avatar is equipped with an averted gaze. Our results suggest that avatar presentation and nonverbal cues should be carefully considered in chatbot interface design to facilitate users' self-disclosure. Implications and applications are discussed.