This 4-week RCT (n=102 Swedish adults with SAD) compared smartphone-based AI-delivered psychodynamic therapy (AI-PDT) vs. AI-delivered CBT (AI-CBT) vs. a waitlist control, randomized 1:1:1, with primary outcome of social anxiety severity on the Social Phobia Inventory (SPIN) at post-treatment and 1-month follow-up.
Both active arms produced moderate within-group symptom reductions (AI-PDT: d=0.79; AI-CBT: d=0.72) but neither was significantly better than waitlist at post-treatment (AI-PDT: p=.204, d=0.43; AI-CBT: p=.727, d=0.18), partly because the waitlist itself improved substantially (d=0.54). At 1-month follow-up, AI-PDT became significantly superior to waitlist (d=0.65, p=.038); AI-CBT did not reach significance (d=0.51, p=.111). No significant differences were found between AI-PDT and AI-CBT at either timepoint. Co-occurring ADHD/ASD predicted meaningfully worse outcomes (Δ≈6.6 SPIN points; d=0.76).
- High attrition (~30–35% in active arms) reduced power, especially given the unexpectedly large waitlist improvement. - No objective engagement metrics were logged due to privacy-by-design encryption, limiting dose-response analysis. - No pre-registered statistical analysis plan; sample was largely educated, employed women, limiting generalizability.
AI-guided CBT and psychodynamic therapy apps can produce moderate social anxiety reductions in 4 weeks, but superiority over the passage of time plus weekly monitoring is not clearly established at post-treatment — clinicians should temper expectations and note that neurodivergent patients (ADHD/ASD) benefit significantly less and may need adapted or hybrid support.
Explore related topics
