AI Chatbots for Mental Health Self-Management: Lived Experience-Centered Qualitative Study

Area of Science:

• Artificial Intelligence in Mental Health
• Human-Computer Interaction
• Digital Mental Health Interventions

Background:

• Large language models (LLMs) are increasingly used for mental health support, including depression self-management.
• Rapid deployment risks include inaccurate information, lack of empathy, and inadequate crisis support, especially when user priorities are overlooked.

Purpose of the Study:

• To explore the experiences of individuals with lived experience of depression using an LLM-based chatbot for self-management.
• To identify perceived benefits, limitations, and concerns to inform harm-mitigating design.

Main Methods:

• A GPT-4o-based chatbot (Zenny) simulated depression self-management scenarios.
• 17 individuals with lived experience of depression were interviewed after interacting with the chatbot.
• Qualitative content analysis was applied to interview transcripts, notes, and chat logs.

Main Results:

• Participants prioritized informational accuracy and applicability, expressing concerns about incorrect or vague information.
• Emotional support was valued, but participants recognized limits in machine empathy and desired human connection.
• A personalization-privacy dilemma emerged, with users wanting tailored guidance but withholding sensitive data.

Conclusions:

• Individuals with lived experience of depression seek actionable information, emotional validation with defined limits, and privacy-preserving personalization from LLM chatbots.
• Design strategies should mitigate harms, positioning LLM tools as complements to human support in depression recovery.