Episode 63
ChatGPT for Clinicians — Take the Keys, but First, Can You Drive?
Educators in Medicine,
In this newsletter, we continue our journey through the fundamentals of AI, its applications in medicine, and its transformative role in faculty development and education. Let’s dive into learning.
Last week, a big OpenAI announcement that, depending on how you read it, was either the biggest thing to happen to point-of-care AI this year or the most aggressive customer-acquisition play in healthcare history: ChatGPT for Clinicians — free, NPI-verified, on tap, for every U.S. doc, NP, PA, pharmacist, and psychologist who can prove they are licensed.
Next was in JGME from Carl Preiksaitis — yes, same Preiksaitis from Episode 62 — with a title that reads less like a journal article and more like a small, polite warning shot: “Deciding How Much to Trust AI for Teaching and Assessment.”
One says: here, take the keys.
The other asks: do you actually know how to drive this thing yet?
🩺 The What and the How — ChatGPT for Clinicians, Demystified
The What. ChatGPT for Clinicians is a dedicated workspace built on GPT-5.4, OpenAI’s most capable model for healthcare, given to verified U.S. clinicians for free. Not a discount. Not a 30-day trial. Free, full-fat, with higher rate limits than the consumer free tier and a few features the rest of ChatGPT doesn’t have.
What they claim is inside the workspace:
Trusted clinical search with real-time, cited answers grounded in peer-reviewed studies, clinical guidelines, and public health sources
Skills — repeatable workflows you can configure once and reuse forever (think: referral letters, prior authorization templates, discharge instructions) and I believe trying to keep up with OpenEvidence on this too
Deep research — multi-source literature reviews compiled in minutes with full citations
Higher GPT-5.4 limits for the long, gnarly, multi-turn conversations that real clinical questions actually look like
A privacy posture above the consumer tier: conversations in the workspace are not used to train models
The How. You go to chatgpt.com/plans/clinicians, sign in (or create an account), and verify through a third-party identity provider using your NPI number and your active license. That’s it. The verification feels less like applying for a DEA registration and more like setting up a Doximity account — a few minutes, no fax machine, no notarized anything.
What’s in it for you? Let me put it bluntly. If you were going to pay for ChatGPT Plus anyway ($20/month) to do the kinds of things doctors quietly already do — draft a patient letter, summarize a guideline, brainstorm a differential — OpenAI just handed that back to you, plus better features, for typing in a number you already have memorized. You get the better model, the higher limits, the citations, and the privacy posture. The cost of entry is your NPI and a few seconds. But, as my brother in law says “there’s no free lunch”. They’re making a play here with NPI’s.
The cynic in me says: of course it’s free — they want every clinician in America logging in every day so the next round of healthcare contracts writes itself. The pragmatist in me says: that’s true and my residents now have access to a better tool than they had last week. Both can be true.
How Should We Use This With Learners?
First, the detail that matters: eligibility is tied to a verifiable license and an NPI. That means most PGY-1+ residents qualify (they hold state licenses; many institutions issue NPIs early in training). It also means most medical students do not — they don’t yet have NPIs, and even if they do, the license check will fail. Worth thinking about: we now have a tier of AI tooling that’s available to your interns but not to your M3s on the same service.
So how do we actually use this with the learners who do qualify?
⚖️ The Trust Question
Now to the question I get asked the most: is this better than what I’m already using?
There are too many tools. I’m not sure yet. How much should we trust it?
The editorial frames trust as something we have to calibrate, not assign. We don’t trust the workspace because OpenAI told us a benchmark number. We trust it the way we trust a medical student on the team: provisionally, in defined domains, with verification, and with our own judgment as the final authority. Trust is earned task by task. Some tasks (drafting a patient letter? sure) are low-trust-required. Others (deciding whether this hypotensive patient gets fluids or pressors? not yet, not without you in the loop).
Some non-negotiable fine print before you start:
This is not a medical device. OpenAI says it explicitly: not for diagnosis or treatment. Outputs require clinician oversight, full stop.
The free clinician workspace is not covered by a BAA. Do not paste PHI. If your institution needs HIPAA coverage for AI, that lives in OpenAI’s healthcare/enterprise track, not the free tier.
We gotta keep asking but how much do you actually trust it, and what are you teaching your learners about that question?
We still need to teach the next generation how to drive — eyes on the road, hands on the wheel, even with the advent of self driving cars.
💌 As always, thanks for reading. Get in touch and let me know your thoughts!
Thank you for joining us on this adventure. Stay tuned for more AI insights, best practices, and more future editions of AI+MedEd.
For education and innovation,
Karim
Share this with someone - have them sign up here.