Open Collections

Abstract

Within the ever-evolving technological landscape, AI-based chatbots have emerged as a potential form of mental health support that may carry less perceived stigma than traditional therapy, are easily accessible, and readily available for both clinical and non-clinical populations (Rackoff et al., 2025). Despite a surge in chatbot development, empirical validation and insight into how individual differences shape user outcomes remains limited. The present mixed methods study examined the efficacy of a rule-based mental health chatbot, Wysa, and explored how individual differences impact well-being outcomes and experiences compared to a comparison group. Specifically, it examined the impact of personality traits and pre-existing attitudes toward AI on perceived stress, negative affect, and positive affect. Additionally, perceptions of relational competencies and therapeutic alliance were explored. Undergraduate students at the University of British Columbia, Okanagan campus (N = 144) were randomly assigned to use Wysa or a journalling comparison app for ten days. Quantitative results indicated that Wysa users reported significantly higher positive affect and lower stress compared to the comparison group. Several significant moderation effects emerged: lower Extraversion predicted greater increases in positive affect for Wysa users, while lower Agreeableness and higher Conscientiousness predicted greater reductions in negative affect and stress for those in the Wysa condition. Baseline attitudes toward AI also significantly moderated perceptions of competencies, such that participants with more positive attitudes perceived Wysa as demonstrating more core competencies (e.g., respect, validation, and non-judgement). Qualitative reports supported these results, with participants describing some emotional benefits, but also noting major limitations in relational aspects (e.g., repetitiveness, lack of understanding, robotic responses). Collectively, these findings suggest that tailoring chatbot content and delivery based on user characteristics may be important to optimizing their impact in mental health contexts.