AI Versus Human-Delivered Online Cognitive Behavioral Therapy for Anxiety Symptoms in Young Adults: A Randomized Controlled Trial

Objective: This study aimed to compare the effectiveness of online cognitive behavioral therapy (CBT) delivered by an AI chatbot versus human peer counselors (participants were told it was AI) in reducing anxiety symptoms in young adults. Methods: Ninety young adults with mild-to-severe anxiety were randomized to a 4-week intervention of AI-CBT (n = 30), peer-counselor-CBT (n = 30), or a no-intervention control (n = 30). The primary outcome, anxiety, was assessed at baseline, mid-point, and post-intervention. Secondary outcomes (the self-efficacy for exercise, sleep quality), psychotherapy benefit, and qualitative user experiences were also evaluated. Results: Both AI and human-delivered interventions led to significant within-group reductions in anxiety (p < 0.05). However, in the primary intention-to-treat analysis, neither intervention demonstrated a statistically significant advantage over the no-intervention control group at post-intervention. A secondary per-protocol analysis suggested a benefit for the human-delivered intervention among study completers. Notably, participants in the AI group reported significantly lower perceived treatment benefit than the human group (p < 0.001). Qualitative analyses indicated that while AI was valued for accessibility and consistency, human intervention was perceived as more flexible in guidance, individualized, emotionally supportive, and conducive to deeper exploration. Conclusions: In this exploratory trial, both AI- and peer-counselor-CBT showed within-group promise, but the evidence does not support their efficacy over a no-intervention control. The AI’s limitations in providing flexible, emotionally supportive, and personalized interaction likely explain the efficacy gap observed between the two interventions. While AI may serve as a scalable support tool, claims of clinical efficacy require significant caution. These preliminary findings warrant replication in a prospectively registered confirmatory trial.