OpenAI makes ChatGPT free for clinicians — but is that enough?

OpenAI just dropped something interesting: free ChatGPT access for verified U.S. physicians, nurse practitioners, and pharmacists. No subscription, no credit card — just a professional license and a pulse. The pitch is that it supports clinical care, documentation, and research.

Let’s be real: this is a smart business move. Healthcare is a massive, high-stakes market, and getting clinicians hooked early on your tool is a long play. But as someone who’s watched AI in healthcare stumble more than once, I have mixed feelings.

The free tier is generous — it includes GPT-4 access, which is the model that actually matters for anything beyond casual chit-chat. Clinicians can ask it to summarize patient histories, draft referral letters, or even help with differential diagnoses. The documentation angle is especially appealing: anyone who’s spent hours writing notes after a 12-hour shift knows how painful that process is.

But here’s the catch: accuracy. I’ve tested GPT-4 on medical questions, and while it’s better than earlier models, it still hallucinates. It’ll confidently list a rare disease that doesn’t exist or cite a study that was never published. For a clinician making a real decision, that’s not just annoying — it’s dangerous. OpenAI is clearly aware of this, which is why they’re not marketing it as a diagnostic tool. But users will push it that way anyway.

Privacy is another elephant in the room. Patient data is protected under HIPAA, and sending it through OpenAI’s servers — even with promises of encryption — makes some hospital compliance officers nervous. OpenAI says they won’t train on API data from healthcare customers, but the free tier might not have the same guarantees. If I were a clinician, I’d want that in writing before I typed a single patient name.

The research angle is less controversial. Using ChatGPT to summarize literature, draft grant proposals, or brainstorm study designs is lower risk. I’ve done it myself for non-clinical work, and it saves real hours. But again, you can’t blindly trust the citations.

What I find most interesting is the timing. Other AI companies like Google (Med-PaLM) and Microsoft (Nuance DAX) are already deep in healthcare, but they’re selling to hospitals, not individual clinicians. OpenAI is going direct-to-doctor, which is a different strategy. It’s faster, cheaper to scale, and bypasses the slow enterprise sales cycle. Smart.

Still, I worry about over-reliance. I’ve seen residents paste entire patient cases into ChatGPT and copy the output verbatim. That’s not using AI as a tool — that’s abdicating judgment. The best clinicians will use this as a second opinion, not a replacement for their own brain.

Overall, this is a positive step. Free access lowers the barrier to experimentation, and that’s how real workflows get improved. But OpenAI needs to invest as much in safety and transparency as they do in marketing. Otherwise, this becomes another cautionary tale about AI in medicine.