From Real-World Context to Design Specifications: Operationalizing Insights for Mental Health Chatbots

Young people's mental health represents a pressing societal crisis, with many increasingly adopting large language model (LLM)-based systems for emotional support. However, the design and development of these systems often fail to take a human-centric approach that surfaces the personalized requirements and safety boundaries of young people, especially those from marginalized communities. Using a three-stage, multi-method participatory design approach, we analyzed global survey data (N = 6,272), conducted three co-creation workshops with marginalized young people (N = 24, Mage = 17.6), and performed semi-structured interviews with community stakeholders (N = 14; parents and social workers). Thematic analysis distilled diverse stakeholder perspectives into actionable design specifications using the Persuasive Systems Design (PSD) framework. Our findings show how firsthand accounts can be operationalized into concrete design principles, foundational for future research on aligning LLMs with the needs of young people and their communities to build safer, more effective digital well-being tools.