Safety, efficacy and acceptability of human-GenAI single-session exposure-based intervention for academic anxiety: randomized controlled trials

Key Findings

• — Across two RCTs with 425 participants, effect sizes on primary outcomes were d = 0.25 to 0.35 against the active control and d = 0.43 to 0.58 against waitlist.
• — Primary outcomes were immediate state anxiety and academic anxiety and procrastination at two-week follow-up; secondary outcomes (positive and negative affect, generalized anxiety, depressive symptoms) showed limited effects.
• — Participants perceived the human-GenAI exposure program as safe and acceptable, which the authors present as initial evidence that GenAI can deliver exposure-based intervention with a human in the loop.
• — Study 1 was preregistered in the Chinese Clinical Trial Registry (ChiCTR2500101784, 2025-04-29); Study 2 was preregistered on OSF (2025-08-13) and registered retrospectively in ChiCTR (2026-04-30).

Methodology Notes

Two randomized controlled trials with waitlist and active control arms, total N = 425, Chinese university students, single-session hybrid human-GenAI exposure program targeting academic anxiety, two-week follow-up. Published open access 2026-09-12 (dc.date on the nature.com landing page, which was fetched directly with a browser user agent). The authors declare a patent application on dialog control in LLM-based psychological interventions. The underlying model is not named in the abstract.

Sources