Understanding Attitudes and Trust of Generative AI Chatbots for Social Anxiety Support

TL;DR

This study investigates how people perceive and trust GenAI chatbots as a tool for social anxiety (SA) support, addressing the problem of accessible mental health aid amid rising SA prevalence. It employs a mixed-methods approach (survey with n=159 and interviews with n=17) to quantify trust across six dimensions and link it to willingness to use, while exploring how SA severity and prior GenAI exposure shape these dynamics. Key findings show a strong link between trust and willingness, with severe SA users prioritizing emotional trust and milder SA users prioritizing cognitive trust; prior therapy experiences influence perceived utility, including a view that GenAI can outperform low-quality psychotherapy. The results yield design and ethical implications for GenAI in SA care, highlighting the need to cultivate emotional engagement alongside technical reliability, and to address autonomy, safety, and evolving client–therapist–GenAI relationships for real-world deployment.

Abstract

Social anxiety (SA) has become increasingly prevalent. Traditional coping strategies often face accessibility challenges. Generative AI (GenAI), known for their knowledgeable and conversational capabilities, are emerging as alternative tools for mental well-being. With the increased integration of GenAI, it is important to examine individuals' attitudes and trust in GenAI chatbots' support for SA. Through a mixed-method approach that involved surveys (n = 159) and interviews (n = 17), we found that individuals with severe symptoms tended to trust and embrace GenAI chatbots more readily, valuing their non-judgmental support and perceived emotional comprehension. However, those with milder symptoms prioritized technical reliability. We identified factors influencing trust, such as GenAI chatbots' ability to generate empathetic responses and its context-sensitive limitations, which were particularly important among individuals with SA. We also discuss the design implications and use of GenAI chatbots in fostering cognitive and emotional trust, with practical and design considerations.

Figures (6)

• Figure 1: Correlation matrix displaying the coefficients between various aspects of trust, including competence, honesty, experience, benevolence, reliability, and expectation) in GenAI chatbots.
• Figure 2: The Kruskal-Wallis test assessed differences in trust levels across willingness groups, with a non-parametric pairwise comparison using Dunn's test identifying specific differences between the groups.
• Figure 3: Participants' willingness to use GenAI chatbots for SA support in relation to (A) symptom severity, (B) length of prior use, and (C) frequency of use GenAI chatbots in coping with SA.
• Figure 4: Clustering using the Gaussian Mixture Model with distinct color-coded clusters (i.e., each color represents a cluster), showing the relationship between (A) duration of GenAI chatbots use and frequency of GenAI chatbots use for SA support, and (B) duration of GenAI chatbots use and severity of SA symptoms. Participants in the Undecided willingness group can be described using a tree structure. The first division separates participants into two groups: those with high SA symptoms who have used GenAI chatbots for SA , and those with low SA symptoms who have never used GenAI chatbots for this purpose. Within the second group, participants are further categorized by the duration of their GenAI chatbots usage, with longer usage and shorter usage . The cluster structure based on the severity of the symptoms suggests that individuals with similar symptom levels show distinct patterns in their engagement with GenAI chatbots.
• Figure 5: Cluster and component definition criteria: (A) Within-Sum-of-Squares (WSS) for GMM clustering, (B) Average Silhouette Method for GMM clustering, and (C) Parallel analysis for determining the number of components to retain in factor analysis.