Generative AI as a bounded elicitation and evidence-synthesis layer for pain assessment in older adults with cognitive impairment

Pain assessment in older adults with cognitive impairment remains challenging, particularly when changes in communication make self-report difficult. Validated observational tools, caregiver and staff knowledge, clinical examination, and repeated within-person assessment remain the foundation of care. However, information from these sources may be incomplete, conflicting, or documented across different settings. Generative artificial intelligence (AI), including large language models, may help clinicians organize and review this information, but it should not independently detect, diagnose, or determine treatment for pain. In this Perspective article, we propose a person-centered, clinician-governed framework for the bounded use of generative AI in pain assessment. The framework organizes seven testable use cases under three core functions: capacity-sensitive elicitation, source-attributed synthesis, and longitudinal support and communication. Potential applications include adaptive pain interviews, identification of possible pain-related information in everyday language, structuring of behavioral observations, multi-informant synthesis, synthesis of validated nonverbal indicators, longitudinal pain-related diaries, and caregiver communication support. Each application requires clear limits, source identification, communication of uncertainty, consideration of alternative causes of distress, and mandatory clinician verification before clinical action or documentation. We also examine risks related to suggestibility, bias, medication effects, privacy, alert fatigue, workload, and overreliance on AI-generated outputs. A five-step research pathway is proposed that includes co-design, technical testing, clinical validation, evaluation across populations and care settings, and risk-based governance. Generative AI should be adopted only if prospective studies demonstrate meaningful value beyond validated assessment methods and usual clinical workflows without reducing patient choice, replacing clinical judgment, or increasing workload.