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

Key Findings

• — 102 adults with social anxiety disorder were randomised 1:1:1 (34 per arm); 73% female, mean age 41, 30% with a co-occurring ADHD or autism diagnosis.
• — Within-group improvement was d = 0.79 for AI-PDT and d = 0.72 for AI-CBT, but neither active arm was significantly superior to the waitlist at post-treatment (waitlist itself improved, d = 0.54).
• — At one-month follow-up AI-PDT became significantly superior to waitlist (d = 0.65) while AI-CBT did not (d = 0.51); therapeutic alliance was established early and remained stable in both arms.
• — A co-occurring ADHD or autism diagnosis predicted poorer outcomes.
• — Conversations were encrypted locally on participants' devices, which precluded central logging of messages; safety relied on a manual review by a licensed psychologist and no inappropriate responses were reported.

Methodology Notes

Recruitment April 2025 via social media in Sweden; 122 screened, 102 randomised; LSAS-SR of 30 or more required; participants with substantial suicide risk on the C-SSRS self-report or severe depression were excluded. Registered at ClinicalTrials.gov NCT07533812, but the authors state no protocol or statistical analysis plan was publicly preregistered before data collection. Sample of 34 per arm gives 80% power only for moderate-to-large effects. Competing interests: two authors published the self-help book that inspired the therapeutic content. Published online 2026-06-13 in Internet Interventions volume 45 (article 100960), gold open access CC BY 4.0; full text verified via Europe PMC (PMC13292244).

Sources