# OpenAI's Singhal Discusses Clinician AI After U.S. Product Launch

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/36280
Published: 2026-10-21T03:22:17.000Z
Updated: 2026-10-21T03:22:17.000Z
Section: Technology

> ChatGPT for Clinicians became available to verified U.S. medical professionals in April. Separate Kenyan studies of AI Consult measured clinical errors and treatment failure.

OpenAI health AI research lead Karan Singhal discussed AI tools for clinicians in an Oct. 20 interview, months after the company introduced a free product for verified U.S. medical professionals.

ChatGPT for Clinicians became available April 22, 2026, to verified individual physicians, nurse practitioners, physician assistants and pharmacists in the United States. It helps clinicians research medical questions and prepare documentation. Clinicians remain responsible for patient care decisions.

Before the release, physician advisers evaluated 6,924 conversations and judged 99.6% of the responses safe and accurate. That review assessed the tool’s answers, not patient outcomes.

Separate research into clinical decision support examined AI Consult, a different tool used at Penda Health clinics in Kenya. Its findings do not measure the performance of ChatGPT for Clinicians.

An initial evaluation included 39,849 visits at 15 clinics. Physicians assessed records and clinical decisions from 5,666 randomly selected visits for diagnostic and treatment errors. Among clinicians with access to AI Consult, diagnostic errors were 16% lower and treatment errors were 13% lower than among those without access. Those relative comparisons concerned errors identified through review of documented care, not patient outcomes.

A subsequent randomized trial of AI Consult 2.0 enrolled 9,691 patients at 16 Kenyan facilities. Its primary analysis included 9,347 encounters and measured treatment failure within 14 days. The outcome occurred in 102 of 4,693 patients, or 2.2%, in the AI group and 94 of 4,654 patients, or 2.0%, in the control group. The difference was not statistically significant.

The studies addressed different questions. The earlier evaluation assessed errors in clinical records and decisions, while the trial measured treatment failure after care. Neither tested whether AI shortened the time between publication of medical research and its routine use.

That research-to-practice interval is often described with a 17-year estimate. The figure is not a universal timeline for updating clinical guidelines: the interval varies by specialty and by how adoption is measured.