Generative Confidants: How do People Experience Trust in Emotional Support from Generative AI?

People are increasingly turning to generative AI (e.g., ChatGPT, Gemini, Copilot) for emotional support and companionship. While trust is central to these unsupervised interactions, we lack an understanding of how people develop and experience trust in this high-stakes context, where inappropriate reliance could delay professional help-seeking or lead to the acceptance of harmful advice. We recruited 24 frequent users of generative AI for emotional support and conducted a qualitative study combining diary entries, chat transcripts, and in-depth interviews. Our results reveal that trust develops through iterative processes: confirming content credibility, learning to shape AI responses, and forming nuanced "artificial human" mental models that balance conversational fluency with awareness of machine limitations. Critically, the consistently personalised, positive, and persuasive language of generative AIs simultaneously builds trust while undermining the critical engagement necessary for appropriate trust. Our findings have immediate implications for AI developers seeking to facilitate rather than discourage critical thinking, mental health professionals navigating clients’ concurrent AI use, and policymakers establishing safeguards for vulnerable populations.