Designing for GenAI-mediated mental health peer support: Lessons from human peer support interactions in a wellbeing app

Mental health services remain difficult to access worldwide, and many young people turn instead to friends or online communities for support, particularly where mental illness is highly stigmatized. We study how such peer support is practised, and what this implies for the design of Generative AI (GenAI). We deployed a peer-support app with 28 Japanese university students over two weeks and qualitatively analyzed their peer-to-peer chat conversations. We identify seven recurring relational patterns, including tentative openings, routes into depth, layered accounts of lived experience, reciprocity, and careful navigation of medical disclosures. Building on these findings, we distinguish two roles for GenAI in this ecosystem: GenAI as support for human peers (backstage tools that scaffold and strengthen human-to-human support) and GenAI as a peer (a conversational companion in its own right). For each, we outline opportunities, limitations, and ethical boundaries, advancing HCI discourse beyond therapist-oriented GenAI, toward more reciprocal, peer-based mental health support.