Key points are not available for this paper at this time.
Abstract Suicide remains a leading cause of preventable death worldwide, requiring timely and scalable interventions. This systematic literature review examines how Artificial Intelligence (AI) has been applied to suicide. Following PRISMA guidelines and a registered PROSPERO protocol, a comprehensive search across APA PsycNET, PubMed, IEEE Xplore, and Scopus yielded 1,293 records. No publication date limits were applied; all eligible studies available up to the final search date (May 2025) were included. After screening and quality appraisal, 160 studies published in peer-reviewed, Q1-ranked journals were included for in-depth synthesis. The review follows an AI taxonomy categorising the different AI technologies into machine learning (ML), deep learning (DL), natural language processing (NLP), generative AI (GenAI), large language models (LLMs), and explainable AI (XAI). It organises findings into several thematic domains, such as social media-based, electronic health records, demographic modelling, clinical transitions, and emerging technologies. The findings revealed that NLP and DL approaches, particularly on social media and clinical datasets, performed better than traditional statistical methods in identifying suicidal behaviour. Population-specific models (by age, gender, and veteran status) enhance prediction accuracy. XAI methods such as SHAP and LIME improve model transparency and clinical trust, while GenAI and LLMs are emerging as promising yet underexplored tools. Despite growing interest, the review identified limitations in cross-cultural generalisability, lack of prospective validation, and underrepresentation of low- and middle-income countries (LMICs). The findings also showed that AI complements, rather than replaces, traditional suicide assessment tools, offering hybrid potential for real-time and personalised suicide prevention. The study concludes with a call for ethically aligned, explainable, and context-sensitive AI frameworks to ensure fair, unbiased, scalable deployment in mental health care.
Mokheleli et al. (Mon,) studied this question.