SleepPathfinder: A Socratic Questioning and Self-Decision-Based Chatbot to Support User Engagement in Digital CBT-I: Usability and Feasibility Study.

Background: Chronic insomnia is a highly prevalent sleep disorder that adversely affects quality of life and mental health. Cognitive behavioral therapy for insomnia (CBT-I) is internationally recommended as the first-line treatment, and digital CBT-I (dCBT-I) has been developed to improve accessibility and scalability. While existing dCBT-I systems effectively support structured behavioral training through standardized protocols, they provide relatively limited support for users' cognitive exploration and meaning-making processes, particularly in helping users reflect on and internalize the rationale behind CBT-I practices in daily life. These limitations may contribute to challenges in sustained engagement and long-term adherence. Objective: This study aimed to examine the usability and feasibility of SleepPathfinder, a conversational CBT-I support chatbot that integrates Socratic questioning and a self-decision mechanism to support users' understanding of and engagement with CBT-I practices. Methods: SleepPathfinder was designed around a 4-stage conversational flow: education on CBT-I techniques, Socratic cognitive exploration, self-decision, and advice provision. We conducted (1) a single-session pilot usability study (n=45) to assess system stability and user experience and (2) a 5-day condition-based comparative experiment (n=30) consisting of daily sessions, comparing an exploratory dialogue condition with a directive, protocol-guided dialogue condition. Quantitative measures assessed usability, cognitive appraisals related to sleep problems, autonomy-related experiences, and behavioral readiness, while qualitative feedback and conversational log analyses were used to examine interaction patterns and engagement characteristics. Results: In the comparative experiment, the exploratory dialogue condition showed a tendency toward reduced perceived threat and severity appraisal of sleep problems compared with the directive condition, accompanied by moderate effect sizes in cognitive perception measures. Autonomy-related experiences, including perceived choice and engagement, demonstrated suggestive upward trends in the exploratory condition. Behavioral intention changes were comparable across conditions, while overall readiness for change increased across participants. Conversational log analyses indicated that greater depth and volume of user self-narrative were associated with larger shifts in cognitive appraisals, whereas the frequency of chatbot questions alone was not. The pilot usability study indicated generally positive evaluations of system usability and content credibility, while identifying areas for improvement in emotional responsiveness and conversational naturalness. Conclusions: These findings suggest that a Socratic questioning-based and self-decision-based conversational structure is usable and feasible as a supportive interaction layer within dCBT-I systems. Rather than altering the directive behavioral structure of CBT-I, such an approach may complement existing protocols by facilitating cognitive exploration and supporting user-perceived autonomy. This study provides design-oriented evidence to inform the refinement of dialogue-supported digital CBT-I systems aimed at enhancing user engagement with CBT-I practices.