Published in NEJM AI, this review article examines how large language models (LLMs) can improve the clinical research informed consent process, which is often limited by long, complex, and difficult-to-understand consent forms. We mapped the emerging use of LLMs across seven major domains: glossary generation, plain-language rewriting, visual aid generation, accessibility and formatting, translation, teach-back and comprehension validation, and chatbot-assisted consent. The article highlights how AI systems could help make consent materials clearer, more accessible, and more engaging for participants while supporting research teams in preparing participant-facing content.
At the same time, the review emphasizes that LLMs must be integrated carefully into ethical and regulatory workflows. Because consent requires accuracy, transparency, and participant understanding, AI-generated outputs introduce risks such as hallucinations, omissions, biased language, translation errors, and uncertainty around review-board oversight. We argue that LLM-assisted consent should be grounded in approved source materials, subject to human review, validated across diverse populations, and supported by strong audit and escalation systems. Done thoughtfully, these tools could help move informed consent closer to its ethical goal: ensuring that participants truly understand the research they are being asked to join.