Majority of Doctors Adopt AI in Clinical Practice

A cluster of surveys from Ireland, the U.S., and national studies show AI moving from experimentation into routine clinical workflows. In Ireland, a joint EY and Royal College of Physicians of Ireland (RCPI) survey of 516 physicians found 58% used AI in the prior year and more than 20% used it daily. In the U.S., Doximity's 2026 State of AI in Medicine report finds 63% of physicians currently using AI and 94% either using or interested. Patient-facing research from West Health shows over 66 million Americans have consulted AI for health information, reinforcing two-sided adoption.

Clinicians report applying Generative AI across clinical and administrative tasks. Common use cases include:

• •suggesting differential diagnoses
• •identifying treatment options
• •generating post-consultation documentation
• •summarizing patient histories and communications

These are largely natural language tasks that map to large language model capabilities: retrieval, summarization, and synthesis. Adoption patterns vary by specialty and age cohort; surveys show higher uptake in neurology, gastroenterology, internal medicine, and among younger clinicians, though Irish data flags strong use in the 50-64 age group as well. Key operational concerns are accuracy, auditability, data governance, legal risk, and clinician over-dependence.

This is not a one-off trend. Multiple independent datasets now show clinician adoption accelerating across geographies and care settings. The pattern mirrors earlier digital transitions: tools that reduce documentation burden and speed information synthesis rapidly find uptake. However, unlike benign productivity apps, clinical AI directly affects diagnosis and treatment decisions and patient privacy. The simultaneous rise in patient-initiated AI use increases the pressure on clinicians to validate and contextualize AI outputs. The net result is a stronger need for three institutional capabilities: rigorous evaluation pipelines, explicit governance and liability frameworks, and scalable clinician training.

Short term, expect more health systems to pilot model governance frameworks and certify discrete workflows where AI delivers measurable time savings or error reduction. Medium term, regulators and payers will drive requirements for audit logs, model performance metrics in situ, and documentation standards linking AI outputs to clinician decisions. For ML practitioners, prioritize evaluation datasets that reflect real-world clinical documentation, calibration on rare conditions, and explainability mechanisms that support clinical reasoning.

Clinicians and ML teams should treat deployment as an integration challenge, not purely a model accuracy problem. Invest in user workflows, human-in-the-loop safeguards, logging and monitoring, and focused training programs. Hospitals need to quantify harms and benefits in their own data and prepare for legal and privacy oversight as adoption scales.