Trauma-Informed Conversational Agents for Mental Health: Understanding User Perspectives and Experiences (Preprint)

BACKGROUND Mental health Conversational Agents (MHCAs), or chatbots, are increasingly used to provide accessible, scalable support for individuals experiencing psychological distress. While these tools, whether AI-driven or scripted, hold promise for expanding mental health care, concerns remain regarding their safety, responsiveness, and appropriateness, especially for trauma-exposed users. Trauma-informed care (TIC), a well-established framework in in-person therapy, emphasizes safety, trust, empowerment, collaboration, peer support, and cultural sensitivity. However, little is understood about how users interpret and prioritize TIC principles when engaging with chatbot-based mental health support. OBJECTIVE This study aimed to explore 1) how users conceptualize trauma-informed care in the context of mental health chatbots and 2) what factors can help us to predict whether users perceive their chatbot interaction as trauma-informed. METHODS A web-based, self-administered survey (REDCap) was completed by 606 participants recruited via ClickWorker, social media, and MyChart. The 59-item survey (multiple response questions, Likert Scale, open-ended) assessed demographics, technology use, adverse life experiences, chatbot use, and perceptions of TIC and satisfaction with MHCAs. To identify latent domains representing how participants structure their understanding of TIC in conversational agents' contexts, Exploratory and Confirmatory Factor Analyses were conducted. To identify factors to predict trauma-informed interaction, multivariable logistic regression models were used. RESULTS Participants reported high overall satisfaction (84.5%) and high trauma-informed perception (92.9%) of their chatbot experience. Factor analyses yielded a validated five-domain structure: Trust & Transparency, Data Safety, Empowerment, Peer Support, and Cultural Sensitivity, with excellent model fit (CFI = 0.978, RMSEA = 0.045). Regression analyses identified Trust (OR = 3.89, p = .001), Empowerment (OR = 1.97, p = .025), and Peer Support (OR = 1.73, p = .021) as significant positive predictors of TIC perception. In contrast, citing convenience as a primary use reason (OR = 0.33, p =.041) and higher smartphone proficiency (OR = 0.38, p = .033) were negatively associated with TIC perception. CONCLUSIONS This study offers an empirical foundation for defining and measuring trauma-informed design in mental health conversational agents. Key TIC features such as trust-building, emotional validation, and peer connection significantly shape user perceptions. Results underscore the need for AI-based mental health tools to align with trauma-informed principles and adapt to diverse user expectations to enhance therapeutic safety, credibility, and inclusivity.