JMIR Publications

ABSTRACT

Conversational artificial intelligence (AI), including general-purpose chatbots and AI companions, has become a common source of information and emotional support for adolescents and young adults, and a growing evidence base indicates that these interactions can intersect with suicide risk. Recent nationally representative data indicate that roughly one in five US adolescents and young adults has used AI chatbots for mental health advice, with the majority not disclosing this use to any clinician. In a nationally representative survey, roughly one in seven adolescent users reported encountering self-harm content from these tools, and a similar proportion suicidal content. At the same time, comparative evaluations of large language models show that these systems can exhibit a systematic upward bias, rating responses to suicidal ideation as more appropriate than expert clinicians do, even as their safety behavior continues to evolve. Psychiatric practice nonetheless lacks an established approach to assessing a patient’s AI exposure. Drawing on an illustrative case in which a young man’s overdose was informed by information obtained from a conversational AI chatbot, we argue that AI exposure merits consideration within routine suicide-risk assessment as one component of a patient’s digital environment history, alongside substance use and access to means. We propose a provisional, practice-oriented framework, summarized by the mnemonic I-CARE (Inquire, Clarify, Assess, Rate, Engage), for incorporating AI exposure into risk assessment and documenting it there. We frame I-CARE not as a validated instrument but as a structured starting point for the kind of rapid, repeatable testing this fast-moving field requires. The aim is practical: to make a clinically relevant exposure something clinicians ask about during assessment rather than something noticed only afterward. A prospective evaluation within our healthcare system is planned.