Toward Generative Experience for Stress Management Interventions

Digital mental health tools are widely used to provide accessible stress support, yet most deliver interventions through fixed interaction formats such as text chat or static exercises. Even when systems can identify appropriate strategies, they remain constrained by the kinds of experiences they are able to enact. This limitation may reduce how well stress management interventions fit users’ immediate needs and contexts. We will build and evaluate a system that adaptively generates user interfaces to enact a personalized stress management intervention. Following a guided user-AI conversation about the user’s stress context, the system will generate multiple candidate intervention experiences that differ in structure, sequencing, and modality use. Candidates will be evaluated using theory-driven rubrics, and the selected experience will be assembled at runtime using interactive multimodal components such as audio guidance, visual scaffolds, timers, short videos, and lightweight interactive elements. In a single session study, we will compare this generative experience condition to a fixed format LLM-based cognitive restructuring activity, examining perceived stress reduction and user experience.