Tailored Virtual Agent Guidance for Stress Management: Comparing Directive and Non-Directive Approaches by Alexithymia Status

Stress significantly affects young adults, highlighting the need for effective digital mental health interventions (DMHIs). However, many DMHIs face high dropout rates and low engagement, partly due to alexithymia, characterized by difficulties in identifying and expressing emotions. Directive guidance support individuals with alexithymia, however, DMHIs rely on one-size-fits-all strategies. This study investigated a large language model (LLM)-powered virtual agent delivering Directive or Non-Directive guidance for stress management and compared two guidance styles according to alexithymia status. The results showed that although stress reduction did not differ significantly across the two guidance styles, individuals with alexithymia reported enhanced therapeutic alliance under Directive guidance, while those without alexithymia formed stronger alliances with Non-Directive guidance, particularly for tasks (p = .03) and goals (p = .05). The study advances understanding tailored guidance by integrating LLM-powered virtual agent, demonstrating that personalizing interventions for alexithymia can strengthen therapeutic alliance and boost long-term engagement in DMHI.