Randomized controlled trial of GenAI coach feedback efficacy, acceptability and mechanisms in anxiety worry logs

Abstract

Abstract Anxiety disorders are prevalent worldwide, yet access to evidence-based interventions remains limited. Digital worry-log interventions are scalable, but unguided use may limit engagement and efficacy, while human-delivered feedback may improve outcomes, yet is labor-intensive and cost-prohibitive. Combining worry-log interventions with GenAI-generated feedback represents a potential solution. We conducted a three-arm randomized controlled trial (registered in the Chinese Clinical Trial Registry, ChiCTR2500101780, April 29, 2025) to evaluate the efficacy, acceptability, and potential mechanisms of GenAI-generated personalized feedback during a 10-day anxiety-focused worry-log intervention, compared with human-delivered feedback and a worry-log-only control condition. Results indicated that the GenAI-generated feedback condition significantly improved state anxiety and worry relative to the worry-log-only control, with improvements and participant-rated acceptability comparable to the human-delivered feedback condition. These findings support using GenAI-generated personalized feedback in a worry-log intervention, offering a scalable approach to expanding access to efficacious anxiety interventions and alleviating the global treatment gap.