Artificial intelligence has rapidly evolved from a back-office efficiency tool to a frontline clinical companion for U.S. physicians, according to new survey data from Doximity. The medical professional network polled more than 3,300 physicians and found that AI adoption has surged across clinical workflows, with daily usage now exceeding 50% and clinical decision support adoption jumping from 37% to 63% in just one year—a 70% increase that signals a fundamental shift in how doctors access and apply medical knowledge.

Yet this accelerated integration comes with significant growing pains. Three-quarters of physicians report encountering at least one factual inaccuracy while using AI for work in the past year, and more than half cite these inaccuracies as the primary reason for reducing their AI use.

AI Moves from Administrative Tasks to Clinical Frontlines

The survey reveals that physicians are increasingly turning to AI for core clinical functions rather than just administrative relief. Seventy percent now use AI to search medical literature, up from just over 50% in 2024, while 63% employ it for clinical decision support—a near-doubling from the previous year's 37%.

More than half of physicians use AI daily, with an additional 33% engaging with these tools at least weekly. The educational impact appears substantial: 81% of physicians report that AI taught them something new, whether updated guidelines, alternative diagnoses, or treatment options. Sixty-five percent say AI tools have actively helped develop their clinical skills and knowledge.

Accuracy Concerns and Skill Erosion Create Adoption Headwinds

Despite widespread adoption, physicians harbor serious reservations about AI's reliability and long-term impact on clinical competence. The 75% error encounter rate represents a critical trust barrier, with more than 50% of doctors identifying inaccuracies as the main factor driving reduced AI use.

Nearly 50% of physicians express concern about overreliance on AI, while more than 25% cite skill erosion as the primary limitation on their AI usage. These concerns extend beyond individual practice: 61% of physicians worry that residents, fellows, and trainees will become overly dependent on AI and fail to develop independent clinical reasoning skills—a concern with profound implications for medical education and long-term care quality.

The Verification Burden Shows Signs of Easing

One potentially encouraging trend: over 25% of physicians report that the effort required to correct or double-check AI output has decreased compared to one year ago. This suggests that either AI accuracy is improving, physicians are developing more efficient verification workflows, or both—though the persistently high rate of encountered inaccuracies indicates substantial room for improvement remains.

The Doximity data paints a picture of a profession in transition, rapidly integrating AI into clinical workflows while simultaneously grappling with fundamental questions about accuracy, autonomy, and the preservation of clinical expertise. As AI becomes a daily fixture in physician workflows, the challenge for healthcare systems and technology developers will be ensuring these tools enhance rather than erode the clinical judgment that remains central to quality patient care.

Sources

  1. Despite concerns, AI becomes daily fixture in physician workflows — Healthcare Dive