Health / Technology

Add as a preferred source on Google

OpenAI Launches Clinician AI as Kenya Studies Show Mixed Results

A study of Penda Health clinics found fewer documented errors, while a later trial of the separate AI Consult 2.0 system found no significant change in 14-day treatment failure.

Updated 2 min read
A clinical officer reviews a patient record during a consultation in Nairobi, Kenya. / TokenPost.ai (macro)
A clinical officer reviews a patient record during a consultation in Nairobi, Kenya. / TokenPost.ai (macro)

OpenAI’s clinician-focused ChatGPT product is free to verified U.S. medical professionals, while separate Kenyan studies of another AI system found fewer record-review errors but no significant drop in treatment failure.

Karan Singhal, OpenAI’s health AI research lead, discussed clinician-focused AI in an Oct. 20 podcast interview. The Kenyan research assessed clinical errors and patient treatment failure, not how quickly medical findings enter routine practice.

OpenAI introduced ChatGPT for Clinicians on April 22, 2026. The service supports medical research and documentation, and clinicians remain responsible for care decisions. U.S. physicians and nurse practitioners, along with pharmacists and physician assistants, can use it at no charge after verification.

Before release, physician advisers reviewed 6,924 conversations with the product. OpenAI said they rated 99.6% of its answers safe and accurate. The review assessed responses rather than patient outcomes.

A separate evaluation by OpenAI and Penda Health covered 39,849 visits at 15 Kenyan clinics. Physicians reviewed a random sample of 5,666 records for diagnostic and treatment errors. In comparisons involving clinicians who had access to AI Consult, the reviewed records showed relative reductions of 16% in diagnostic errors and 13% in treatment errors. Those figures describe errors identified in records, not treatment outcomes for patients.

A later cluster-randomized trial of AI Consult 2.0 assigned clinical officers in groups at 16 Kenyan facilities. It enrolled 9,691 patients, with 9,347 encounters included in the primary analysis. Within 14 days, treatment failure was recorded for 102 of 4,693 patients in the AI group (2.2%) and 94 of 4,654 in the control group (2.0%). The researchers found no statistically significant difference. The trial’s results were published in Nature Medicine on June 26, 2026.

OpenAI said at the product launch that it planned to pilot access for verified clinicians in other countries through the Better Evidence Network, subject to local regulations. The company described the pilot as planned for the coming months but set no fixed decision date.

Related articles

View all