New medical research can take somewhere between 5 and 17 years to show up in the clinical guidelines doctors actually follow. OpenAI thinks that window could collapse to days, hours, or even minutes.
That was the pitch from Karan Singhal, who leads health AI research at OpenAI, during an a16z podcast episode on October 20, 2026. A study published today might shape the care of a patient who isn’t born yet. Singhal would like to speed that up.
The lag problem, and OpenAI’s proposed fix
By his count, the gap between a finding and its arrival in guidelines and physician behavior runs anywhere from 5 to 17 years.
OpenAI’s answer leans on its general-purpose AI models and a product called ChatGPT for Clinicians. The tool is designed to turn dense medical literature into practical guidance a doctor can use. The goal is to get that translation done within days, hours, or minutes.
OpenAI launched ChatGPT for Clinicians in April 2026. It gives verified US clinicians access to advanced models along with HIPAA options. Before launch, the product went through testing involving approximately 7,000 conversations. OpenAI reported a 99.6% safety and accuracy rating from that round.
HealthBench and the physician feedback loop
OpenAI built HealthBench as a benchmark for evaluating how AI handles clinical situations. It relied on input from 262 physicians, who wrote rubric criteria for judging responses to realistic clinical scenarios.
HealthBench uses 48,562 physician-written criteria to evaluate 5,000 realistic conversations. The physicians involved came from 26 specialties and 60 countries.
Singhal stressed that this kind of physician-sourced data matters during model training, not just after. His view is that close collaboration with doctors is what keeps AI tools aligned with what healthcare professionals actually need right now.
Early results from the field
OpenAI has pointed to a study with Penda Health as early evidence. Across 39,849 patient visits in July 2025, the study reported a 16% relative reduction in diagnostic errors. It also reported a 13% decrease in treatment errors.
Singhal also cited a figure that frames the whole conversation. More than 300 million people use ChatGPT for health and wellness questions every week.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.