AI Meets Mental Health: What Happens When ChatGPT Writes CBM-I Materials?



Che-Wei (Jerry) HSU

1The University of Otago, Department of Psychological Medicine, NZ

2 Otago Polytechnic, College of Community Development and Personal Wellbeing, NZ

CBM-Interpretation targets negative interpretation bias, but developing materials is resource-intensive. Here, we examined whether AI could support CBM-I development. Study1: ChatGPT rated human- and AI-generated assessment scenarios; although results wereinconclusive, ratings aligned with intended bias direction and were more precise for AI materials. Study2: participants with self-reported depression rated AI and human CBM-I scenarios as different but directionally consistent, with greater AI precision. Study3: both AI- and human-generated CBM-I interventions reduced negative interpretation bias relative to controls, with non-inferiority between AI and human materials. Findings suggest AI may support CBM-I development to promote scalability and personalization of training materials.