"I Can Simply Be Myself": Co-Designing Conversational Agents for Advance Care Planning

Abstract

Advance directives preserve autonomy, but they involve abstract, challenging choices. Existing decision support for advance directives structures decisions rather than facilitating the articulation of preferences. This limitation points to large language model–based conversational agents (CAs) as a potential alternative. We conducted a three-phase co-design study with a panel of six older adults. This involved multiple conversations with CAs, group discussions, and collaborative prompting. Although participants expressed ambivalent views on whether CAs are suitable, they did offer recommendations on how they should be designed. They saw the CA as a level-headed advisor that provides a space to talk through choices and uncertainties on their own, without immediate pressure from others’ expectations and feelings. CAs should make preferences visible and summarize them. Moreover, participants wanted the CAs to challenge underlying values to ensure that the decision feels right. CAs become facilitators of sense-making, which provisionally link personal meaning to future outcomes.