What You Should Know
– OpenAI has launched ChatGPT Health, a dedicated, secure environment that allows users to connect their medical records and wellness apps (e.g., Apple Health, MyFitnessPal) directly to ChatGPT’s intelligence.
– Built in collaboration with over 260 physicians and utilizing a new clinical evaluation framework called HealthBench, the platform aims to synthesize scattered health data into actionable insights for the 230 million people who already ask health questions on the platform weekly.
– To enable medical record access, OpenAI partnered with b.well, a leading secure health data network, ensuring that users maintain explicit control over what data is shared.
How ChatGPT Health Works: Architecture and Capabilities
ChatGPT Health operates as a separate environment within the ChatGPT interface, architecturally isolated from standard conversations to maintain privacy boundaries. When users initiate health-related discussions in regular ChatGPT, the system prompts them to move into the Health space to activate additional protections. Once inside Health, conversations, uploaded files, and connected data remain compartmentalized—Health-specific information never flows back into general ChatGPT conversations, though the system can pull limited context from non-Health chats when relevant (such as recent life changes that might affect health decisions).
The separation extends to memory functions. Health maintains its own memory system that retains user preferences, health context, and conversation history separately from general ChatGPT memory. Users can view and delete Health memories independently through dedicated controls, providing granular management over what information the system retains over time.
Core Use Cases supported by ChatGPT Health include:
Understanding Medical Test Results: Users can upload lab reports or connect medical records, then ask ChatGPT Health to explain what specific values mean, how they compare to normal ranges, what factors might influence results, and what questions to ask healthcare providers about findings. The system translates clinical terminology into accessible language while flagging concerning values that warrant immediate clinical attention.
Appointment Preparation: By reviewing medical history, current symptoms, and recent test results, ChatGPT Health helps users formulate comprehensive questions for upcoming doctor visits, ensuring they address all concerns within limited appointment timeframes. The system can generate structured question lists prioritized by clinical urgency based on physician input during development.
Treatment and Medication Guidance: While explicitly not providing diagnosis or treatment recommendations, ChatGPT Health can explain how prescribed medications work, potential side effects to monitor, interaction risks with other drugs, and strategies for managing treatment regimens. This addresses medication adherence challenges that affect an estimated 50% of patients with chronic conditions.
Wellness Planning: Integrating data from wearables and wellness apps, the system offers personalized suggestions for diet modifications, exercise routines, sleep optimization, and stress management tailored to individual health patterns, fitness levels, and medical conditions rather than generic wellness advice.
Insurance Navigation: Users can describe their healthcare utilization patterns and ask ChatGPT Health to compare insurance plan options, explaining coverage differences, out-of-pocket cost implications, and network considerations relevant to their specific health needs—a particularly valuable capability during open enrollment periods when consumers struggle with complex benefit comparisons.
Development Methodology
Over two years, OpenAI engaged more than 260 physicians practicing across 60 countries and representing dozens of specialties to inform Health’s design and response quality. These clinicians provided over 600,000 feedback instances across 30 focus areas, evaluating model outputs for safety, clarity, appropriate urgency in recommending clinical follow-up, and alignment with evidence-based medicine principles.
The physician network reviewed how ChatGPT Health handles diverse scenarios: explaining cardiovascular risk factors, interpreting psychiatric medication side effects, addressing pediatric developmental concerns, navigating cancer treatment options, managing chronic pain, understanding genetic test results, and countless other health questions that consumers commonly research independently. This breadth aims to ensure the system responds appropriately across the full spectrum of health concerns rather than optimizing for narrow use cases.
HealthBench Evaluation Framework
To systematically assess response quality, OpenAI developed HealthBench—an evaluation framework created with physician input that moves beyond traditional medical AI benchmarks. Rather than testing performance on licensing exam questions or academic medical quizzes, HealthBench evaluates responses using physician-written rubrics that reflect real-world clinical judgment priorities.
The framework emphasizes several dimensions:
Safety: Does the response appropriately escalate urgent symptoms? Does it avoid suggesting dangerous self-treatment? Does it account for contraindications and drug interactions?
Clarity: Is medical terminology explained accessibly without oversimplifying? Are probabilities and uncertainties communicated clearly? Can patients with varying health literacy understand the information?
Appropriate Care Escalation: When should the response direct users to emergency care versus urgent care versus routine appointments versus self-management? How urgently should specific symptoms prompt clinical evaluation?
Individual Context Respect: Does the response acknowledge that health decisions depend on personal circumstances, values, medical history, and preferences rather than offering one-size-fits-all recommendations?
This evaluation approach reflects a philosophy that consumer health AI should be judged by how well it supports patient-clinician relationships rather than whether it can pass medical exams—a distinction with important implications for accuracy standards, use case boundaries, and regulatory treatment.
Availability
You can sign up for the waitlist to request access.
