AI Beats Doctors at Coding, 1 in 6 Use Health Chatbots, UK Fast-Tracks Rules

Three key developments are reshaping healthcare AI today, September 26, 2025. These stories show how AI is moving from lab to bedside — and into patients' daily lives.

Have suggestions? Reply to this email.

A Mount Sinai study found a game-changing method for AI medical coding. The technique asks AI to describe a diagnosis, then match it to real examples before picking a code. This simple step made AI more accurate than doctors in many cases.

Researchers tested 9 AI models on 500 emergency visits. Each AI output was matched to 10 similar cases from over 1 million hospital records. Mount Sinai reports that models with the lookup step had higher accuracy and outperformed physician-assigned codes.

This matters because it cuts admin time for clinicians and could speed code suggestions in EHRs. The method works even with small, low-cost models when given example lookups. Health systems can pilot this approach in EHR workflows to lower billing errors and reduce coding workload. NEJM AI

Patient behavior is shifting fast. About 1 in 6 adults now regularly use AI chatbots for health info or advice. Among adults under 30, it's 1 in 4, according to The New York Times.

Patients seek quick answers, symptom checks, and guidance before seeing clinicians. This affects demand patterns, triage needs, and patient expectations. But it also creates liability issues if chatbot advice is wrong or misleading.

Healthcare practices should monitor patient chatbot use during visits. Update triage workflows for self-directed pre-visit queries. Educate patients about vetted tools and clear warnings about limitations. Document when chatbot advice influenced care decisions for risk management.

The UK launched the National Commission on AI Regulation in Healthcare to speed safe AI use across the NHS. National Health Executive reports the commission will advise regulators and publish a new rulebook next year.

The panel includes experts from Google, Microsoft, clinicians, researchers, and patient advocates. Early priorities include guidance on voice technology for clinicians, radiology, pathology, and remote monitoring.

Key stats show AI's current NHS impact: 100% of stroke units in England use AI for diagnosis support, 50% of hospital trusts use AI for detection like lung cancer screening, and hospitals using AI diagnostics report 42% fewer diagnostic errors.

This clears regulatory uncertainty for trusts and vendors while speeding safe deployment of tools that cut admin work and diagnostic errors. Expect clearer rules and faster, regulated adoption next year.

These developments show AI is no longer experimental in healthcare. From coding accuracy to patient self-service to national policy, AI is becoming standard practice. Success will depend on careful implementation, clear governance, and keeping humans in the loop for safety.