ChatGPT for Clinicians goes free in the U.S. as doctor AI use hits 72%

OpenAI is making a bigger bet on vertical AI than the headline “free for U.S. clinicians” suggests. ChatGPT for Clinicians packages frontier models, cited search across peer-reviewed medical sources, reusable workflow skills, and deep research into a workspace aimed at the most time-starved part of U.S. healthcare: documentation, evidence review, and day-to-day clinical decision support.

The timing is not accidental. OpenAI cites an American Medical Association survey showing 72% of physicians now use AI in clinical practice, up from 48% a year earlier. The company says clinician usage of ChatGPT has more than doubled over the past year, which means this is no longer an experimental niche. Hospitals already buying enterprise AI want compliance, while individual doctors want something they can use immediately. OpenAI is trying to cover both: free access for verified U.S. physicians, nurse practitioners, physician assistants, and pharmacists, plus optional HIPAA support through eligible accounts with a Business Associate Agreement.

The feature list shows where the company thinks value actually appears. ChatGPT for Clinicians can return cited answers from millions of peer-reviewed sources, build reusable skills for routine tasks such as referral letters and patient instructions, run literature reviews, and even turn eligible evidence review into continuing medical education credit. These are workflow tools, not generic chat tricks. The source also points to a new open benchmark, HealthBench Professional, built from real clinician chat tasks across care consult, documentation, and medical research.

OpenAI is leaning hard on safety data to make the release credible. Physician advisors have reviewed more than 700,000 model responses, according to the company. Before release, clinicians tested 6,924 conversations drawn from daily work. OpenAI says physicians rated 99.6% of those responses safe and accurate, and on a 355-example citation subset the system referenced ground-truth sources more often than human physicians. Those numbers do not eliminate clinical risk, and OpenAI explicitly says the product is designed to support, not replace, clinician judgment. But they do move the product discussion away from vague promises and toward measured operational use.

The broader implication is that frontier model vendors are no longer content to wait for hospitals to wrap general-purpose chatbots in their own software. They want the clinical workflow layer too. If this rollout works, the next competitive battleground in healthcare AI will not be raw model intelligence. It will be trust, citations, workflow repeatability, and how much clerical time a product can actually remove from a clinician’s week.