Most people already use ChatGPT for health questions. Turns out, over 230 million people ask health-related questions on ChatGPT every single week.
That’s a lot of people turning to an AI tool that was never specifically built to handle the weight of that responsibility. OpenAI clearly took note. Because what they’ve done next is not just a product update. It’s a meaningful rethink of how AI should behave when someone’s health is in the picture.
Here’s what OpenAI actually built, how they built it, and what it means for both clinicians and everyday patients.
The Problem They Were Solving
Healthcare is stretched thin. Demand is rising. Clinicians are overwhelmed by administrative work, and critical medical knowledge is scattered across countless sources. A doctor seeing a patient often has to act on the immediate problem in front of them because there’s simply no time to look at the full picture.
Meanwhile, AI adoption in healthcare was accelerating fast. According to an American Medical Association survey, physician use of AI nearly doubled in a single year, going from 38% in 2023 to 66% in 2024. But many of those doctors were using personal tools because their organizations hadn’t adopted anything formal yet.
OpenAI’s response was to build something specifically designed for the clinical environment, rather than expecting clinicians to adapt a general-purpose chatbot to a professional setting it wasn’t designed for.
What OpenAI Actually Built
OpenAI launched two distinct things, and it’s worth understanding them separately because they serve different people.
1. ChatGPT Health (for patients)
This is a dedicated space inside ChatGPT where you can securely connect your medical records, Apple Health data, and wellness apps like MyFitnessPal and Function. Once connected, ChatGPT Health can help you make sense of recent lab results, prepare questions before a doctor’s visit, interpret data from wearables, and summarize care instructions in plain language.
Here’s the thing that makes this different from just Googling your symptoms: the responses are grounded in your actual health data, not generic population-level information. If you connect your records and ask about a recent blood test result, it’s working with your specific numbers.
On the privacy side, health conversations are encrypted, stored separately from your regular ChatGPT chats, and crucially, they will not be used to train OpenAI’s foundation models. That last point matters more than it might seem at first glance.
2. ChatGPT for Healthcare (for clinical institutions)
This is the enterprise product. It’s already rolling out to institutions like AdventHealth, Baylor Scott & White Health, Boston Children’s Hospital, Cedars-Sinai Medical Center, HCA Healthcare, Memorial Sloan Kettering Cancer Center, Stanford Medicine Children’s Health, and UCSF.
For clinical teams, the product pulls from peer-reviewed literature, clinical guidelines, and institution-specific pathway documents to generate context-aware responses with citations. It can draft discharge notes, referral letters, and prior authorization documents. It supports role-based access, audit logs, customer-managed encryption keys, and HIPAA-compliant infrastructure.
In short: it’s built for the regulations and workflows that actually exist inside healthcare organizations, not just built for people who happen to work in healthcare.
The Part That Changes Everything: How OpenAI Trained These Models
This is the part that genuinely sets this apart from previous AI health tools, and it’s worth slowing down here.
Over two years, OpenAI worked with more than 260 licensed physicians across 60 countries and dozens of specialties. That physician network reviewed more than 600,000 model outputs across 30 areas of clinical focus.
That’s not a small advisory board signing off on a product. That’s an ongoing, iterative feedback loop that directly shaped model training, safety mitigations, and how the product actually behaves. The physicians guided decisions on when to escalate to urgent care, how to communicate medical information clearly without oversimplifying it, and how to handle sensitive moments where the stakes are genuinely high.
They also put the model through multiple rounds of physician-led red teaming, specifically testing for failure modes before the product went live.
What Is HealthBench and Why Should You Care?
One of the most important things to come out of this work is something called HealthBench, an open evaluation framework that OpenAI built with that same physician network.
Most AI benchmarks in healthcare test for factual recall. Think: can the model pass a medical licensing exam? HealthBench does something more interesting. It evaluates responses against 5,000 realistic health conversations, each graded using physician-written rubrics that reflect how actual clinicians judge quality in real practice.
The rubrics assess four things:
- Safety: Does the response flag urgent symptoms appropriately? Does it avoid suggesting dangerous self-treatment?
- Clarity: Is medical information explained accessibly, without stripping away the nuance that matters?
- Appropriate escalation: When should the response send someone to the ER versus urgent care versus a scheduled appointment?
- Individual context: Does it treat health decisions as personal, not one-size-fits-all?
What this really means is that OpenAI is measuring what good clinical communication actually looks like, not just whether the model can recall facts. GPT-5.2, the model powering these products, outperforms earlier models and comparator systems on HealthBench’s most challenging clinical workflow evaluations. It also outperforms human baselines across every role measured in a separate evaluation called GDPval.
HealthBench is open-source, which means other organizations can use it to evaluate their own health AI models. That’s a meaningful contribution to the broader field, not just to OpenAI’s own products.
How to Get Access Right Now
If you want to try ChatGPT Health as a patient or general user, here’s exactly what to do:
- Go to chatgpt.com and log in to your account
- Look for the Health option in the left sidebar menu
- If you don’t see it yet, join the waitlist here
- Once inside Health, you can connect your medical records and wellness apps
- Start by asking it to help you understand a recent test result, or prepare questions for your next appointment
If you’re part of a healthcare institution and want to explore ChatGPT for Healthcare at an organizational level, OpenAI has an enterprise pathway. You’d want to reach out directly through the OpenAI for Healthcare page.
The Honest Caveats
OpenAI has been clear that ChatGPT Health is not a replacement for professional medical care. It is not designed for diagnosis or treatment. If you’re in a medical emergency, call emergency services, don’t open an app.
There’s also a real limitation worth knowing: like any data stored in the cloud, information in ChatGPT Health could still be obtained through a subpoena or court order. OpenAI’s privacy protections are strong, but they’re not absolute. That’s worth understanding before you share sensitive health information.
The hallucination problem in AI hasn’t disappeared either. These models are significantly more reliable than previous versions, but they can still produce inaccurate information. The physician-led training and HealthBench evaluation reduce this risk meaningfully, but they don’t eliminate it. Cross-checking anything important with your actual doctor remains the right move.
What This Actually Signals
The competitive dynamic here is worth noting. Anthropic has also launched Claude for Healthcare. Google has been pushing Gemini into clinical workflows. The big AI labs have all spotted the same gap: healthcare is one of the highest-stakes, highest-volume use cases for AI, and it’s been underserved by general-purpose tools.
What OpenAI has done differently is invest two years and 600,000+ physician feedback instances into the model before going to market with a dedicated health product. That’s a deliberate signal: that getting this right required more than repackaging an existing model with a healthcare label on it.
The real test, as always, will be in real-world use over time. But the foundation looks more solid than anything that’s come before it.
Have you tried ChatGPT Health yet? I’d genuinely love to hear what your experience has been. Drop a comment below or find me on BlueSky.
Read the full OpenAI announcement here: Making ChatGPT Better for Clinicians
Just Being Resourceful! 
Leave a comment