Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance: A Randomized Clinical Trial
Among distressed university students, a conversational artificial intelligence program reduced anxiety and improved well-being more than group therapy or a wait list.
*Randomized clinical trial; Level 1b (OCEBM).
*Randomized clinical trial; Level 1b (OCEBM).
Citation
Shoshani A, Gurfinkel B, Kor A, et al. Efficacy of a conversational artificial intelligence agent for psychiatric symptoms and digital therapeutic alliance: a randomized clinical trial. JAMA Network Open. 2026;9(4):e266713. doi:10.1001/jamanetworkopen.2026.6713
Background
Scalable mental health support is needed where clinician time is limited. Conversational artificial intelligence tools may increase access, but comparative evidence and how “working relationship” affects outcomes are limited.
Patients
995 university students (Israel), ages 18-35, reporting current distress. Excluded: active suicidal thoughts or crisis, current psychotherapy or psychiatric medication, and history of severe mental disorders.
Intervention
12-week conversational artificial intelligence mental health platform with on-demand messaging and coping exercises.
Control
Face-to-face group therapy (12 weekly 90-minute sessions) or wait list.
Outcome
Anxiety (primary), depression (primary), posttraumatic stress symptoms (primary), well-being (primary), life satisfaction (primary); plus engagement and perceived working relationship with the tool.
Follow-up Period
12 weeks (primary endpoint) plus 3 months.
Results
| Outcome | Comparison | Adjusted mean difference (95% confidence interval) |
|---|---|---|
| Anxiety score (primary) | Conversational artificial intelligence vs group therapy | -2.17 (-2.67 to -1.67) |
| Anxiety score (primary) | Conversational artificial intelligence vs wait list | -2.15 (-2.65 to -1.65) |
| Depression score (primary) | Conversational artificial intelligence vs wait list | -1.99 (-2.63 to -1.35) |
| Well-being score (primary) | Conversational artificial intelligence vs group therapy | +5.72 (+2.71 to +8.73) |
| Well-being score (primary) | Conversational artificial intelligence vs wait list | +9.16 (+6.14 to +12.18) |
| Life satisfaction (primary) | Conversational artificial intelligence vs wait list | +2.58 (+1.62 to +3.54) |
Posttraumatic stress symptoms: no meaningful differences between groups.
Clinical improvement (moved from above to below common symptom cutoffs at 12 weeks): anxiety—57.9% with conversational artificial intelligence vs 9.8% wait list (number needed to treat 2); depression—47.0% vs 14.0% (number needed to treat 4).
Analyses were intention-to-treat. No serious adverse events were reported.
Limitations
Students with severe illness, suicidality, and current treatment were excluded, limiting generalizability. Outcomes were self-reported and participants were unblinded. Benefits were modest and not seen for posttraumatic stress symptoms. Follow-up loss was about 35%. Some authors had financial ties to the platform company.
Funding
Not reported; some authors employed by or held equity in the platform company.
Clinical Application
Consider conversational artificial intelligence as an adjunct or early support for mild-to-moderate distress; do not rely on it for trauma-related symptoms or higher-risk patients.
Discussion
Sign in to join the discussion.
In this 3-arm randomized clinical trial of distressed university students, a 12‑week conversational AI platform improved anxiety vs face-to-face group therapy (GAD-7 mean difference −2.17, 95% CI −2.67 to −1.67) at 12 weeks; how should we judge internal validity (unblinded, self-report, unequal “dose”) and decide whether/where to implement this in routine care?