Artificial Intelligence in Suicide Prevention: A Systematic Review of Randomized Controlled Trials on Risk Prediction, Fully Automated Interventions, and AI-Guided Treatment Allocation

Background: Artificial intelligence (AI) has been proposed as a transformative tool in suicide prevention, yet most evidence remains observational. To provide a rigorous benchmark, we systematically reviewed randomized controlled trials (RCTs) evaluating AI-based interventions targeting suicidal thoughts, behaviours, or help-seeking. Methods: Following PRISMA 2020 guidelines, MEDLINE, Web of Science, and Scopus were searched to 31 May 2025. Eligible studies were RCTs in humans that incorporated AI or machine learning for risk prediction, automated intervention, or treatment allocation. Methodological quality was assessed with the PEDro scale and certainty of evidence with GRADE. Results: From 1101 screened records, six RCTs (n = 793) met all criteria. Three studies tested machine learning risk prediction, two evaluated fully automated interventions (a transformer-based recommender and a digital nudge), and one examined AI-assisted treatment allocation. Risk-prediction models stratified short-term suicidal outcomes with accuracies of up to 0.67 and AUC values around 0.70. Digital interventions reduced counsellor response latency or increased crisis-service uptake by 23%. Algorithm-guided allocation reduced the occurrence of suicidal events when randomisation aligned with model recommendations. Methodological quality was moderate to high (median PEDro = 8/10), but GRADE certainty was low due to small samples and imprecision. Conclusions: AI can enhance discrete processes in suicide prevention, including risk stratification, help-seeking, and personalized treatment. However, the current evidence is limited, and larger multisite RCTs with longer follow-up, CONSORT-AI compliance, and equity-focused design are urgently required.