Informing the design of question-asking conversational agents for reflection

Reflecting on everyday experiences offers valuable insights and has the potential to enhance psychological well-being. Yet, only some have access to a facilitator for reflection. Conversational agents hold promise as companions for these discussions. We surveyed individuals with therapy experience to understand user needs and arrived at interaction strategies used in therapy. We then evaluated these strategies with five therapists and transformed our data, along with their input, into a set of interaction strategies to be used on conversational agents for reflection. We developed an AI chatbot prototype where we implemented these strategies and conducted a 1-week in-the-wild study with 34 participants to evaluate the interaction strategies and experiences of interacting with a chatbot for reflection. Findings reveal that participants are willing to engage with a chatbot, even with limited capabilities. Critical aspects include the chatbot’s contextual awareness, statement repetition, and human-like qualities. Successfully balancing questions with non-question statements is essential for a pleasurable dialogue-driven reflection. Our paper presents implications for future design and research studies.