72% of U.S. doctors now use AI for work.
Last year, that figure was 48%. A jump of 24 percentage points in one year—faster than any consulting firm predicted.
The problem? Most doctors are using the general-purpose version of ChatGPT, not one designed for clinical scenarios, systematically validated by medical professionals, or connected to authoritative medical literature.
On April 23, OpenAI launched ChatGPT for Clinicians, a version specifically for doctors.
This isn't just a skin change—it's a substantial difference.
Who can use it, and is it free?
In the U.S., verified physicians, nurse practitioners (NPs), physician assistants (PAs), and licensed pharmacists can use it for free. Not a trial, not a feature-limited free version—the full GPT-5.4-based model.
Why free?
OpenAI's reasoning: AI penetration in this industry is already high, but most are using general-purpose tools. Rather than letting doctors work on unverified general ChatGPT, offer a specialized version. The implicit logic: build presence first, then sell the HIPAA-compliant Business Associate Agreement (BAA) version to enterprises.
That business logic makes sense.
How rigorous was the development?
700,000+—that's the number of model responses reviewed by OpenAI's physician advisors.
Not internal testing—real doctors sat down and reviewed them one by one.
The final result: 99.6% of responses were rated safe and accurate, tested on nearly 7,000 clinical dialogues. That's higher than most expected.
Another interesting metric: HealthBench Professional. This is a new benchmark OpenAI released alongside, specifically measuring AI capability in clinical scenarios. ChatGPT for Clinicians scored 59.0—outperforming "human doctors with ample time and web search access."
What does "doctors with time and search access" mean? The most favorable human conditions—sitting at a computer, able to look up literature, not rushed. The AI still won.
What features does it have?
Trusted Clinical Search
Real-time retrieval of millions of authoritative peer-reviewed articles with citations. Doctors can ask "what's the latest evidence for this medication regimen" and get sourced literature conclusions, not general search results.
Reusable Workflow Skills
A practical design. Doctors define common repetitive tasks as skills—like "write a referral letter," "write pre-operative consent," "fill out prior authorization"—and trigger them to have ChatGPT follow the same steps and format each time.
Standardizing repetitive work is exactly what consumes the most of doctors' daily time.
"This tool addresses the physician burnout crisis with a low barrier to entry."
— OpenAI release notes
Automatic CME Credit Tracking
U.S. licensed doctors need to complete continuing medical education (CME) credits regularly. ChatGPT for Clinicians can automatically record literature reviews to help track credits.
A small feature, but very practical for clinical work.
Privacy Protection
Key point: Doctor conversations will not be used to train OpenAI's models.
For general ChatGPT users, this has always been a gray area. For the clinical version, OpenAI is clear: no training, HIPAA-compliant agreements supported. Essential for any medical scenario involving patient information.
Why "this time is different"
Over the past three years, AI in healthcare has been a tug-of-war between "hype vs. deployment." Various AI diagnostic systems and imaging analysis tools mostly stayed in hospital pilots.
This time, the logic is different—not trying to replace doctors' diagnostic decisions, but focusing on solving administrative and information retrieval inefficiencies.
Writing referral letters is high-value work done by high-value doctors. Searching literature is necessary but time-consuming. Filling prior authorizations is necessary but worse. These tasks don't require clinical intuition but consume massive amounts of doctor time daily.
"The system is explicitly designed to assist, not replace, the judgment and expertise of clinicians."
— OpenAI release notes
The meaning behind the HealthBench Professional 59.0 score: AI surpasses humans in information processing and synthesis, but that doesn't mean AI can make clinical judgments. OpenAI itself hasn't claimed that.
The boundary is clearly drawn.
What's next?
Now the free version is running in the U.S. Next steps likely include:
- Expansion to other countries (regulatory compliance is the main barrier)
- Enterprise version (hospital/health system level, with deeper EHR integration)
- After data accumulation, fine-tuned versions for specific specialties
72% of U.S. doctors are already using AI. Behind that number is a reality: the penetration war for healthcare AI is nearly over. What's next is competing to migrate these users from general tools to specialized versions.
That's the real meaning behind this free strategy.
Sources: Making ChatGPT better for clinicians (OpenAI official release, April 23, 2026); OpenAI launches ChatGPT for Clinicians to streamline medical workflows (Neowin); CocoLoop; OpenAI launches ChatGPT for Clinicians: 6 things to know (Becker's Hospital Review)