Microsoft Copilot Health Conversations Analysis Reveals Personal Medical Queries
A research project examining more than 500,000 anonymized health-related dialogues with Microsoft Copilot from January 2026 uncovers notable trends in the application of conversational AI for medical questions. The researchers created a hierarchical intent taxonomy comprising 12 categories, which was confirmed through expert evaluation. Topic-clustering driven by LLMs highlighted common themes. Among the findings, nearly 20% of discussions pertained to personal symptom evaluation, while 40% fell under general information, particularly regarding specific treatments, suggesting a minimum level of personal health interest. Alarmingly, one in seven personal health inquiries exhibited troubling patterns. The study employed LLM-based classification and analysis of usage trends. Insights into health information methods through conversational AI are presented in this research, published as arXiv:2604.15331v1.
Key facts
- Study analyzed over 500,000 de-identified health conversations with Microsoft Copilot
- Research period covered January 2026
- Developed hierarchical intent taxonomy with 12 primary categories
- Used privacy-preserving LLM-based classification validated by human experts
- Applied LLM-driven topic-clustering for prevalent themes
- Nearly 20% of conversations involved personal symptom assessment or condition discussion
- 40% of queries fell into general information category focused on treatments and conditions
- One in seven personal health queries showed concerning patterns
Entities
Institutions
- Microsoft