Abstract
As sleep health becomes increasingly central to personal well-being, individuals are turning to digital tools for education, tracking, and behavior change support. Large language models like ChatGPT and Gemini have recently emerged as promising components of these tools, capable of generating personalized, conversational, and context-aware sleep guidance. This scoping survey synthesizes findings from 21 papers that explore the use of large language models in nonclinical, everyday user-focused applications for sleep health. We organize the literature into 4 core use cases: educational question answering, condition-specific support (eg, obstructive sleep apnea), personalized recommendations and coaching, and cognitive behavioral therapy-based self-help systems. We analyze the diverse data sources involved-including wearable sensor data, self-reported metrics, and synthetic benchmarks-as well as model architectures, fine-tuning techniques, and personalization strategies. Finally, we examine evaluation frameworks ranging from expert review to pilot user studies and LLM-based scoring. The review highlights current capabilities, methodological challenges, and open opportunities for advancing trustworthy, personalized sleep support using generative artificial intelligence, while emphasizing that much of the evidence remains preliminary, often short-term, expert-rated, or proxy-based, which limits external validity and generalizability.