Anthropomorphism in AI Mental Health Chatbots : Impact on Trust and Engagement

Sammanfattning

As artificial intelligence (AI) chatbots increasingly support mental health interventions, understanding how anthropomorphism influences user experiences is critical. This mixed-methods study investigated how different levels ofanthropomorphism in AI chatbots affect trust, emotional comfort, engagement, and perceived human-likeness among users with mild mental health challenges. Sixteen participants (aged 22-40) were randomly assigned to interact with one of three chatbot personas: Orion (low anthropomorphism), Lyra (mediumanthropomorphism), or Nova (high anthropomorphism). Using a between-subjects design, participants completed asynchronous surveys via Google Forms, including Likert-scale items and open-ended questions. Quantitative data were analyzed for descriptive statistics (e.g. means, standard deviations, and median) to assess trustand comfort across conditions, while qualitative responses underwent thematic analysis to identify themes and patterns. Findings suggest that a medium level ofanthropomorphism (Lyra) elicited higher trust and comfort compared to low or high anthropomorphism, with qualitative themes highlighting emotional connection and perceived empathy. These results underscore the importance of balanced anthropomorphism in designing AI companions for mental health support, offering insights to developers and practitioners to enhance user engagement and therapeutic outcomes.