Leveraging Participatory Personas for Reflexive Co-design of Personalization in Large Language Models

Abstract

Large language model (LLM)-based conversational systems are increasingly proposed as personalized digital well-being tools (DWTs) for youth, yet current personalization approaches often fail to reflect heterogeneous lived experiences. We present a participatory approach for eliciting personalization requirements for LLM-based DWTs, with youth well-being as a case study. In a multi-stage co-creation study with youth, parents, and youth care professionals (N=38), we combined data-driven personas with participatory persona co-creation and follow-up stakeholder interviews to examine what meaningful LLM-based DWT personalization should entail. Our preliminary findings suggest that stakeholders prioritize person-centered personalization beyond demographic or clinical labels, emphasizing identity, relational dynamics, routines, literacy needs, and changing lived context. Participants also highlighted the importance of dialogic mechanisms, particularly reflective questioning and ongoing updating of contextual representations over time. We argue that participatory personas can serve as reflexive scaffolds for surfacing community-grounded personalization needs and translating them into emerging design requirements for adaptive LLM systems. These findings position personalization in DWTs as an ongoing dialogic, person-centered process that warrants further investigation in research on human-AI personalization.