The Roles of Verbal Socialness and Visual Anthropomorphism in Mental Health Chatbot Counselors for Individuals with Social Anxiety

Abstract

As artificial intelligence technology advances, mental health chatbots gradually becomes a prevalent tool for counseling and intervention. The chatbot-assisted psychological counseling provides a new avenue for socially anxious individuals who may hesitate to seek help from human counselors. Social anxiety, or social phobia, is characterized by the fear of negative evaluation and the distress in social interaction situations. This research examines whether chatbots could potentially alleviate these symptoms, and improve their felt support and willingness to follow mental health suggestions.

It is a debatable question related to the inclusion of verbal socialness in communication style for mental health counseling. As for machine interlocutors, it also has always been a long-standing question as to what communication style (social-oriented vs. task-oriented) these non-human communication partners should use. In addition to communication style, the integration of anthropomorphic visual cues is a frequently-used strategy to improve the interaction between chatbots and humans. However, the combination effect of the anthropomorphism and communication style remained unclear in chatbots for mental health counselling. It is imperative to examine the combined impact of anthropomorphism and communication style. For those with social anxiety, the effects of communication style and anthropomorphism are more complex.

Drawing upon the ‘Computers Are Social Actors’ (CASA) paradigm and cognitive models of social anxiety, I designed and conducted two empirical studies. These studies aim to examine the influences of communication style and anthropomorphism in mental health chatbots on (1) users’ mind perceptions towards chatbots, (2) the relational experience with the chatbot (i.e., perceived intimacy), (3) emotional response (i.e., fear of negative evaluation), (4) informational response (i.e., perceived information overload) and (5) interaction outcomes (i.e., adherence to suggestions and felt support).

Study 1 was an online experiment with a 2 (communication style: social-oriented vs. task-oriented) × 2 (social anxiety: low vs. high) × 2 (anthropomorphism: low vs. high) design. The chatbot used in this study asked participants about their current moods and provided self-care suggestions in the interaction process. The results show a three-way interaction of communication style, social anxiety and anthropomorphism on mind perception towards the chatbot. For individuals who were more socially anxious, communication style and anthropomorphism had a significant interaction effect on mind perception. However, such an interaction effect did not exist for people with a low level of social anxiety. The following analysis using structural equation modeling further showed that mind perception positively predicted perceived intimacy, fear of negative evaluation, and perceived information overload. Perceived intimacy was linked to increased suggestion adherence and felt support, while fear of negative evaluation negatively affected felt support. Perceived information overload was negatively associated with suggestion adherence.

Study 2 was conducted with a 2 (communication style: social-oriented vs. task-oriented) × 2 (anthropomorphism: low vs. high) online experiment design among people who have a relative high level of social anxiety. The chatbot in Study 2 was deigned to specifically provide knowledge and coping strategies for social anxiety. The objective was to validate the previous study’s results and to explore deeper into the effectiveness of mental health chatbots in mitigating social anxiety symptoms. The results also showed a significant interaction effect. The positive effect of social-oriented communication style on mind perception was strengthened by anthropomorphism. Consistent with Study 1, mind perception positively predicted perceived intimacy, fear of negative evaluation and perceived information overload. Perceived intimacy positively predicted felt support and suggestion adherence. Perceived information overload negatively predicted felt support and suggestion adherence. These studies contribute to the understanding of psychological mechanism underlying interactions with mental health chatbot. Furthermore, they offer practical insights for the development of AI-driven moderation strategies to address social anxiety effectively.

Date of Award	13 Aug 2025
Original language	English
Awarding Institution	City University of Hong Kong
Supervisor	Ki Joon KIM (Supervisor) & Yu-Li LIU (External Co-Supervisor)