A review published in Healthcare Technology Letters in 2026 synthesised 24 cross-sectional studies on patient-facing large language models (LLMs) that provide information about long-term conditions outside clinical settings. The researchers searched MEDLINE, CINAHL, EMBASE and Web of Science between April and May 2025, following PRISMA guidelines, and excluded tools designed to support clinicians.

The studies found that LLMs performed well on general health topics but struggled with specialised information and often produced complex answers with unreliable citations. The review identified concerns about inaccurate information, insufficient empathy and worsening health inequalities. Its analysis of legal challenges focused mainly on liability and consent.

The authors said technical limitations and regulatory gaps around device classification and safety obligations could pose risks to patients. They recommended clear accountability frameworks and said LLMs should serve only as supplementary tools, not replacements for expert clinical advice. They also called for research into agent-based LLM architectures, in which specialised agents collaborate to verify information, reason symbolically and access patient health records under strict data governance.

A review published in Healthcare Technology Letters in 2026 synthesised 24 cross-sectional studies on patient-facing large language models (LLMs) that provide information about long-term conditions outside clinical settings. The researchers searched MEDLINE, CINAHL, EMBASE and Web of Science between April and May 2025, following PRISMA guidelines, and excluded tools designed to support clinicians.

The studies found that LLMs performed well on general health topics but struggled with specialised information and often produced complex answers with unreliable citations. The review identified concerns about inaccurate information, insufficient empathy and worsening health inequalities. Its analysis of legal challenges focused mainly on liability and consent.

The authors said technical limitations and regulatory gaps around device classification and safety obligations could pose risks to patients. They recommended clear accountability frameworks and said LLMs should serve only as supplementary tools, not replacements for expert clinical advice. They also called for research into agent-based LLM architectures, in which specialised agents collaborate to verify information, reason symbolically and access patient health records under strict data governance.