/ Case study

Copilot Health

Helping people make sense of their own health data

At Microsoft AI, I led the product design of Copilot Health, a dedicated health space inside Copilot.

Video

Introducing Copilot Health

Video description: Copilot Health greets Maya with “Hey Maya, ready to get started?” and she types “Why do I feel so tired?” Copilot Health connects information from her record, apps, and wearable: a new blood pressure medication added two weeks ago that could be disrupting her sleep, an irregular sleep schedule due to travel, a heart rate higher than usual at night, and intense workouts close to bedtime. It then lists what may help: review medication timing with a clinician, move intense workouts to the morning, keep a consistent wake-up time, add a 20-minute wind-down routine, and get 10 minutes of outdoor light within an hour of waking, with follow-up suggestions such as questions for a doctor’s appointment. The video ends with “Copilot has clarity,” “Your health. Understood.,” and the Copilot Health logo with “Try at copilot.com/health.”

/ Background

People already have plenty of health information (test results, wearable data, health records) but little help making sense of it. Copilot users field over 50 million health sessions a day, telling us that this is a high-intent use case.

My role was to lead product design for a space where people could bring their records, wearables, and labs together and get answers drawn from credible health organizations, with privacy and safety designed in from the start.

Bar chart of health intent across all conversations: Health Information and Education 40.7%, Research and Academic Support 11.6%, Symptom Questions and Health Concerns 10.9%, Medical Paperwork 9.8%, Fitness, Lifestyle and Coaching 9%, Condition Information and Care Questions 8%, Emotional Wellbeing 3.9%, Healthcare Navigation and Access to Care 3.1%, Insurance, Coverage and Benefits 2.7%, Digital Tools and Fitness Apps 0.4%. General health information dominates, but nearly 1 in 5 conversations involve personal symptom assessment or condition management. — Figure 01
What people already ask about health: intent across all conversations.

Role

Principal Product Design Manager, Microsoft AI

Timeline

2026

A dedicated space for health

Health data lives in its own space inside Copilot rather than as a mode of general chat. This keeps people’s health chats and connected health information secure, encrypted, and private.

Copilot Health does not replace a doctor. Rather, it helps people arrive at appointments prepared with better questions and context.

Welcome to Copilot Health dialog on desktop. It explains that you can ask any health question, prepare for appointments, and get support with wellbeing goals; get more tailored responses by uploading files or connecting wellness apps and health records; and that health conversations, memory, and connected data are stored securely, separate from other chats. Buttons read Get started and Back to Copilot. — Figure 03
Copilot Health sits apart from general Copilot, with a profile that brings connected records, wearables, and labs into one picture.

Discussing Health Concerns, with the note “Your health conversations are private and separate from your other chats.” Below are switches to connect your health records, Apple Health, Oura, and Function. — Figure 03
Copilot Health sits apart from general Copilot, with a profile that brings connected records, wearables, and labs into one picture.

Health sources you can trust

Answers draw on credible health organizations across 50 countries, verified by our clinical team using National Academy of Medicine principles, with clear citations and Harvard Health answer cards.

When people are ready to take the next step, they can search a real-time directory of US providers by specialty, location, language, and insurance.

Video

Cited health answers and finding care in Copilot Health

Video description: Ella asks Copilot “Tell me about the latest acne treatments.” The answer opens with a Harvard Health answer card on acne treatment, then breaks down emerging therapies, common treatments such as salicylic acid washes, benzoyl peroxide gels, azelaic acid, tretinoin, and antibiotics, and special considerations, with cited sources from aad.org and dermatologytimes.com. Ella then asks “Thanks! Could you find a female dermatologist near me in Dallas?” Copilot shows a map of female dermatologists in the Dallas area with provider cards, and opening one shows the doctor’s specialty, contact details, education, accepted insurance, and languages spoken.

Privacy, security, and safety by design

Health conversations and data are isolated from general Copilot, with added access, privacy, and safety controls. Data is encrypted at rest and in transit, protected by strict access controls, and not used to train models.

Product Responsibility Considerations

Data co-mingling

Health conversations are isolated from general Copilot, so what people share about their health stays in Copilot Health.

Consent and connection

People choose which records and devices to connect.

Control and deletion

People can disconnect any source instantly and delete their data at any time.

Model training

Health data is not used to train models.

Overreliance

Copilot Health is positioned as preparation for care with medical disclaimers in context: it is not intended to diagnose, treat, or prevent disease.

Unsafe or inaccurate outputs

An internal clinical team and an external panel of 230+ physicians from 24+ countries review the experience, and new capabilities are released only after clinical evaluation, with clear labeling.

Figure 04

Figure 05

The data boundary: health conversations and connected data stay inside Copilot Health.

General Copilot

Everyday questions, writing, and tasks

General conversations

Agentic workflows and productivity tools

Never crosses the line

Health conversations

Connected records, wearable data, and lab results

Health data for model training

Copilot Health