Study reveals people ask AI chatbots about health most often

The research shows how people rely on AI chatbots for more than simple health facts, from late-night symptom concerns to assistance with appointments and paperwork.

Research: Public launch of a generalist LLM chatbot for health questions. Image credit: Azurhino / Shutterstock

In a recent study published in the journal natural healthMicrosoft researcher A.I.Redmond, Washington, USA, analyzed more than 500,000 anonymized health-related conversations with Microsoft Copilot to identify the characteristics people ask about their health.

Health is one of the high-stakes areas where people ask artificial intelligence (A.I.) Chatbot. Conversational A.I.especially those that utilize large-scale language models (LLMTools like ChatGPT, Copilot, and Gemini are playing an increasingly important role for many users, from being the first point of contact at the onset of symptoms, to asking questions about medications, to understanding interactions with healthcare professionals and the healthcare system. Conversational A.I. This represents a major shift in the way humans interact with digital technologies and information platforms.

Research design of co-pilot health status query

In this study, researchers used Microsoft Copilot to analyze health-related conversations and characterize the questions people ask about their health. A random sample of Copilot conversations was drawn daily in January 2026. Each conversation was assigned a general topic, general intent, and privacy protection summary, and conversations categorized as “health and fitness” were included in this study.

Furthermore, each conversation LLMbased classifier. next, LLMThe based clustering method was applied to a random subsample of 10,000 conversations. Each conversation in this subsample was annotated with additional attributes. of LLM We received summaries and attributes of approximately 250 conversations and grouped them by user journey.

Findings on health information and personal intentions

Overall, the analysis dataset contained 617,827 conversations categorized as health and fitness. The largest health purpose category was health information and education, accounting for approximately 41% of conversations.

This category includes non-personalized health questions, such as general nutritional information, causes of medical conditions, and drug effects. Because some common queries may reflect personal concerns, the actual percentage of personal concerns may be higher, and the reported percentage may be a lower bound.

Additionally, many questions were about specific conditions or treatments rather than general health information, suggesting that people may be seeking general health information for personal decision-making. Most mobile conversations took place at night, while desktop conversations mostly took place during the day. The distribution of health preferences varied significantly across platforms.

Health intent usage distribution (percentage of conversations).

Device type and time of day usage patterns

Usage patterns differed across devices, with the exception of health information and education, which accounted for approximately 40% for both device types. Differences were most pronounced in terms of personal and professional intentions. For example, academic support and research accounted for 16.9% of conversations on desktop, but 5.3% on mobile. Meanwhile, symptom questions and health concerns accounted for 15.9% on mobile compared to 6.9% on desktop.

Stratifying health preferences by time reveals that desktop use of Copilot often occurs alongside other activities such as research, writing papers, and document processing. For example, conversations about medical administration peaked during work hours, while conversations related to academic support and research increased throughout the day, especially after school and work hours. Additionally, personal intentions increased in the evening or night, while academic intentions decreased.

However, the authors noted that these temporal patterns are based on cross-sectional data and may reflect differences in who uses Copilot at different times of the day, as well as within-person variation.

Personal health questions and their impact on care navigation

Finally, the team used a subsample of 2,165 conversations to explore who the health questions were about. This subsample included three personal intentions: psychological well-being, questions about symptoms and health concerns, and questions about condition information and care.

Across all categories, most questions were about personal concerns. However, for the symptom questions and status information categories, 1 in 7 questions were posed by someone else, such as a partner, child, or parent.

Taken together, the findings reveal a clear pattern: A.I. Participating in health-related conversations. Questions about personal health, especially questions about symptoms and mental health, increased in the evening and evening hours. This pattern of questions about happiness is consistent with previous research on the circadian rhythm of negative emotions. Negative emotions tend to be lowest in the morning, increase throughout the day, and peak at night. However, this study could not determine whether this reflects changes in sentiment within the same user or differences between users active at different times.

Nearly one-fifth of conversations involved users describing personal symptoms, test results, or conditions. Additionally, usage patterns vary widely by device type. Personal health purposes were more common on mobile, while desktop use primarily included academic support, medical administration, and research.

Percentage of conversations related to three intents (symptom questions, condition information, and mental health) related to users, dependents, other users, or unknown users.

The survey also found that many users turn to Copilot for help navigating the healthcare system, including finding healthcare providers, understanding coverage, and managing appointments and documents. A.I. is used to deal with administrative frictions and health issues.

This study has some limitations. First, the analysis relied only on Copilot logs that reflect the specific platform and user context.

Second, the sample included conversations over a one-month period. Therefore, seasonal effects can affect the distribution of intents.

Third, because this study only examined questions and not outcomes, we could not determine whether users sought subsequent care or whether the information they received improved their decision-making.

Future research should aim to determine whether information is provided through conversation. A.I. actually help the user.