OpenAI Integrates ChatGPT into Patient Medical Records

OpenAI Introduces ChatGPT into Patient Medical Records

OpenAI has recently launched a new feature for ChatGPT that allows users to connect their health data from Apple Health and their medical records directly to the chatbot. This integration is available for users aged 18 and older, on web and iOS platforms, and for all subscription levels: Free, Go, Plus, and Pro.

Users have the option to link their Apple Health data as well as, when possible, information from U.S. hospital systems such as One Medical or Function Health. Once this data is synchronized, ChatGPT can extract information such as medications, lab results, recent medical visits, sleep data, and activity logs, integrating them into any conversation rather than confining them to a separate section.

OpenAI initially tested this feature with a limited group of users, where it was necessary to open a dedicated area to obtain responses based on their personal data. However, the company observed that over 70% of health-related conversations occurred outside of this dedicated space, often during discussions on topics such as meal planning or unrelated symptoms.

This observation led to a redesign of the feature. Now, ChatGPT can use connected health information in any conversation, provided the user has given consent. For example, a person planning a dinner could receive restaurant recommendations that take into account their recorded dietary restrictions. Similarly, a weekend activity suggestion could be adjusted based on a recent injury noted in the synchronized records.

The Health tab in the application’s sidebar retains its role but has evolved into a management hub where users can connect their accounts, review synchronized data and trends, navigate prompt suggestions, and revisit past health-related discussions.

Testimonials from Early Users of the Health Feature in ChatGPT

OpenAI has gathered numerous testimonials from its early access group, illustrating how the tool is perceived as a support rather than a medical diagnosis.

Blake, a technical program manager, expressed that the most useful aspect of the tool was its ability to transform fragmented medical history into a comprehensible timeline. Having multiple overlapping health issues, Blake found that ChatGPT helped to connect these elements clearly, explain medical terms in simple language, and create shareable summaries for a physical therapist or trainer.

Blake also emphasized that, rather than simply seeing disconnected diagnoses, imaging results, and surgical notes, he could understand the overall pattern of his health. This made the information more usable and provided him with better language to advocate for his interests with providers and trainers.

Reweti, a portfolio manager, highlighted longitudinal analysis as a differentiating factor compared to standard research or a one-off doctor visit. He explained that ChatGPT could access his existing lab results through the connection of all his data, allowing it to infer non-obvious patterns and establish connections he might not have made himself.

However, not all testimonials were without friction. Shannon, a nurse, discovered an unexpected entry in her own record through the tool. She explained that using the Health feature reinforced her belief that one of the greatest strengths of AI is not to replace healthcare professionals but to help patients better understand and navigate their own health information.

This distinction between a tool that highlights something for human follow-up and a tool that makes a clinical decision is crucial for OpenAI as the use of this feature intensifies.

Practical Feedback from Other Testers

Other users emphasized the practical management of data that this feature is intended to simplify. Daniel, a consultant, found that the combined view of health data was more useful than isolated chat sessions. He stated that connecting Apple Health and MyChart made the information more grounded in his daily life.

Kathleen, a small business owner, described how the model adjusted suggestions based on activity gaps it could see directly. She explained that she could ask for suggestions to move, and Health would propose activities suited to her recent activity level.

Carlton, an operations manager, previously used a workaround by exporting spreadsheets from Apple Health to import into ChatGPT. He found that the new integration made the process much smoother and allowed him to view his fitness journey from a different perspective.

OpenAI estimates that health-related queries on ChatGPT exceed 300 million per week, covering various topics from deciphering lab results to preparing for a doctor’s visit. If accurate, this figure places ChatGPT in an enviable position that most digital health platforms would take years to achieve.

The company insists that Health is a support tool and not a diagnostic one, encouraging users to confirm any important information with their healthcare provider.

Performance of OpenAI's AI Models and Their Evaluation

OpenAI attributes the viability of this feature to its new models: GPT‑5.5 Instant, accessible to free users, and GPT‑5.6 Sol, reserved for paid tiers. GPT‑5.5 Instant has improved its ability to recognize situations requiring urgent care and to explain uncertainty, achieving performance comparable to OpenAI's leading models.

GPT‑5.6 Sol is described as the company's highest-performing health model, capable of reasoning through multiple data points such as lab trends over time. To validate these claims, OpenAI collaborated with hundreds of physicians to develop health scenarios and evaluation rubrics.

The models were evaluated on criteria such as accuracy, safety, communication, contextual awareness, completeness, and appropriate escalation to professional care. OpenAI reports that each GPT‑5.6 model outperformed GPT‑5.5 on HealthBench Professional, an internal assessment.

A graph included in the announcement shows that GPT‑5.6 Sol scores higher than GPT‑5.5 Instant and GPT‑4o in categories such as accuracy, communication, and completeness. Physicians also tested the Health product live to assess performance and safety in real-world settings.

Health Data Management and Permission Architecture

Connected health data, along with all resulting conversations, are excluded from the training of base models and advertising targeting, regardless of a user's broader training settings. Access to this data is permission-based, and ChatGPT requests consent before using connected health data to personalize a response.

Users can choose to always allow the use of their data or completely disable prompts. Disconnecting a data source results in its deletion from OpenAI's systems within 30 days, although existing conversations remain until manually deleted.

Memory creation is also regulated. Memories can be created from health conversations but not directly from connected records or Apple Health data. Users can avoid memory creation by using Temporary Chat or by disabling memory in the settings.

OpenAI also reports additional checks for actions that could expose health data via other connected plugins. Red teaming exercises are conducted for these scenarios, although results are not published.

Data Accuracy and Limitations

Data quality remains a point of friction. A medication may remain listed in a patient's record even after they have stopped taking it. OpenAI advises users to report changes directly to ChatGPT and to verify any important information with the original source.

Data from fitness apps and wearable devices depend on what each third-party app chooses to share via Apple Health. Some proprietary scores may not be transferred.

The crucial question is whether the permission model, data deletion timelines, and escalation logic are robust enough to ensure the secure and accurate use of health data.