How people use Copilot for Health

A Microsoft-led research team has published a comprehensive analysis of how people use AI for health queries, examining over 500,000 de-identified conversations with Microsoft Copilot from January 2026. The study, authored by 18 researchers including Mustafa Suleyman and Dominic King, developed a hierarchical intent taxonomy using privacy-preserving LLM-based classification to categorize health conversations into 12 primary categories. Their findings reveal that nearly 20% of all conversations involve personal symptom assessment or condition discussion, suggesting conversational AI is becoming a significant tool for personal health management.

Five key patterns emerged from the data. First, one in seven personal health queries concern someone other than the user—such as children, parents, or partners—indicating AI's role as a caregiving tool. Second, personal queries about symptoms and emotional health increase markedly during evening and nighttime hours when traditional healthcare access is most limited. Third, usage diverges sharply by device: mobile concentrates on personal health concerns while desktop is dominated by professional and academic work. Fourth, 40% of conversations fall under general information but focus on specific treatments and conditions. Fifth, a substantial share of queries involves navigating healthcare systems, including finding providers and understanding insurance, highlighting existing healthcare friction.

The research team used LLM-driven topic-clustering to identify prevalent themes within each intent category, validating their classification against expert human annotation. These patterns have direct implications for platform-specific design, safety considerations, and the responsible development of health AI. The study suggests that current usage patterns represent a lower bound on personal health intent, with AI serving as both a 24/7 health companion and a tool for managing caregiving responsibilities.