In July 2025, TIME magazine reported on a collaboration between OpenAI and Penda Health, a Kenyan primary care provider, which tested the use of an artificial intelligence (AI) assistant called AI Consult to improve clinical decision-making in real-world settings.
The results suggest AI can meaningfully reduce diagnostic and treatment errors in busy, resource-limited environments for primary care doctors.
The project deployed AI Consult, a custom-built generative AI tool based on OpenAI’s GPT-4, across 16 clinics in Nairobi. Over the course of several months, the tool was used to support nearly 40,000 patient visits, offering real-time assistance to clinicians making diagnostic and treatment decisions.
Key findings included:
-
16% reduction in diagnostic errors
-
13% reduction in treatment errors
-
Improved clinician confidence and clinical learning
-
No workflow disruption—the AI operated quietly in the background
AI Consult acted like a clinical co-pilot, analyzing information in real time during patient consultations. It used a traffic-light alert system:
-
Green when care matched best practices
-
Yellow when there was room for improvement
-
Red when the care plan significantly deviated from local guidelines
The tool focused on subtle but common errors, such as inappropriate medication doses, missed diagnoses, or delays in urgent referrals. Importantly, it didn’t replace human judgment but augmented it, giving clinicians evidence-based nudges where needed.
Family doctors and GPs often work under time constraints, making rapid decisions with limited diagnostic tools. This study shows that carefully designed AI tools can:
-
Reduce cognitive burden
-
Serve as on-the-job educational tools
-
Support consistent, evidence-based care—even in overburdened health systems
Moreover, this implementation was done without expensive hardware or disruption to the clinical workflow—highlighting scalability and feasibility even in low- and middle-income countries (LMICs).
Limitations
While the findings are promising, experts have raised valid concerns about the study’s design:
-
60% missing outcome data
-
Unclear inter-rater reliability
-
No peer-reviewed publication (yet)
Reference links —————————
TIME- AI Helps Prevent Medical Errors in Real-World Clinics
OpenAI – Pioneering an AI clinical copilot with Penda Health
