Anna vs. Judith: A randomized comparison of AI-delivered psychodynamic and cognitive behavioral therapies for social anxiety disorder

Artificial intelligence (AI) offers a potential solution to the scalability limits of internet-based psychological interventions. This randomized controlled trial evaluated two smartphone-based, AI-delivered interventions for social anxiety disorder (SAD): Psychodynamic Therapy (AI-PDT) and Cognitive Behavioral Therapy (AI-CBT). One hundred and two adults with SAD were randomized (1:1:1) to AI-PDT, AI-CBT, or a waitlist control for a 4-week daily intervention. The primary outcome was social anxiety severity measured by the Social Phobia Inventory, and analyzed using Linear Mixed Models. Both interventions yielded significant, moderate reductions in symptoms from baseline to post-treatment, with within group effect sizes of d = 0.79 (AI-PDT) and d = 0.72 (AI-CBT), and with no significant differences observed between the two active treatments. However, the between-group difference against the waitlist control at post-treatment was not significant (AI-PDT: p = .204, d = 0.43; AI-CBT: p = .727, d = 0.18), possibly due to substantial improvement in the waitlist condition (d = 0.54). At 1-month follow-up, the difference widened: AI-PDT became significantly superior to waitlist (d = 0.65), while AI-CBT did not (d = 0.51). Therapeutic alliance was established early in both conditions and remained stable throughout treatment, suggesting that AI-delivered interventions can foster a therapeutic alliance regardless of theoretical orientation. Secondary regression analysis revealed that a co-occurring diagnosis of ADHD or autism spectrum disorder significantly predicted poorer outcomes. These findings suggest that the comparative efficacy of AI-guided interventions may increase over time and are capable of fostering moderate symptom reductions, though future iterations require adaptation to better support neurodivergent users.