AI-generated vs. human therapist feedback in internet-based CBT for social anxiety disorder: A six-arm randomized controlled trial with deception design

INTRODUCTION: Guided internet-delivered cognitive behavioral therapy (ICBT) is effective for social anxiety disorder (SAD), but therapist-delivered feedback limits its scalability. Whether AI-generated feedback can match human feedback, and whether patients' expectations about the source shape outcomes is unknown. METHOD: In a six-arm randomized controlled trial, 263 adults with SAD received psychologist-reviewed AI-generated or human feedback. They were told it came from a human or an AI which was correct or incorrect information (four fully crossed feedback arms), or were allocated to unguided self-help or a waitlist. The Social Phobia Inventory (SPIN) was analyzed using linear mixed models, complemented by Bayesian estimation. Secondary outcomes included therapeutic alliance, treatment credibility, negative effects, depression, and anxiety. RESULTS: Active treatment yielded large reductions in social anxiety relative to waitlist (d = -1.23). Before treatment, participants rated it as more credible when they expected feedback from a human rather than AI. AI and human feedback did not however statistically differ in symptom reduction (d = 0.10, 95% CI [-3.62, 5.91]); non-inferiority was not established. Neither the actual source, participants' beliefs, nor their congruence influenced symptoms or alliance. Depression, anxiety, and negative effects did not differ by source. Stronger alliance predicted subsequent symptom reduction, but not the reverse. CONCLUSIONS: Psychologist-reviewed AI-generated feedback yielded reductions in social anxiety symptoms that did not differ significantly from human feedback, without greater harm, and despite participants' initial expectation that AI would be less credible. The barrier to AI-supported ICBT may therefore lie in expectations rather than outcomes. TRIAL REGISTRATION: ISRCTN11343758.