The Department of Veterans Affairs has quietly built one of the largest ambient AI documentation programs in American medicine. Since the agency first switched on AI scribes at 10 medical centers in October 2025, VA primary care teams have used the technology in more than 986,000 appointments, and veterans have declined to have it recording their visit less than 1 percent of the time. Now, following what officials describe as successful pilots, the VA says it will keep expanding the tools nationwide through an enterprise contract in 2026, reaching more specialists and, eventually, select outpatient care providers.
From 10 pilot sites to nearly a million visits
Ambient AI scribes work by listening to the conversation between a clinician and a patient, then automatically drafting a clinical note, transcription, or summary for the electronic health record. The VA’s rollout began with a small pilot footprint 11 months ago and has since scaled to cover all VA Patient Aligned Care Teams’ primary care providers, along with practitioners in behavioral health interdisciplinary programs, physical medicine and rehabilitation, and medical and surgical specialties. Two vendors anchor the deployment: Abridge, whose scribe is live at 70 VA sites and works across both the VA’s legacy VistA system and its newer Oracle-based electronic health record, and Knowtex, which has been used across 10 Veterans Integrated Service Networks and 79 VA medical centers over the past six months.
The numbers behind the pitch
Knowtex’s tool alone has reached more than 7,000 clinician users and, according to the VA, has saved upward of 450,000 hours of EHR documentation time. The company reports an 88 percent adoption rate among clinicians who tried it and an average satisfaction score of 4.48 out of 5 across more than 5,100 ratings. A three-month pilot in Kansas City produced patient-facing numbers as well: veteran satisfaction rose to 95.8 percent, up roughly three points from baseline, while 95.4 percent of veterans said their provider’s explanations were clear, and 95 percent reported trusting their primary care team, both modest but measurable gains attributed in part to clinicians spending less time typing and more time facing the patient.
Background: a documentation crisis the VA didn’t create alone
Clinician burnout tied to EHR documentation burden has been a persistent problem across American healthcare, and the VA, which serves more than 9 million enrolled veterans through one of the country’s largest integrated health systems, has faced it acutely. Ambient scribes emerged over the past two years as the most widely adopted generative-AI application in clinical medicine precisely because they target that pain point directly, converting spoken visits into structured notes without requiring a clinician to learn a new interface mid-appointment. The VA’s approach has leaned on requiring explicit patient consent for every recorded visit, plus a one-time prerequisite training combining ethics instruction and technology setup before any provider can turn the tool on.
The optimistic case from VA leadership
VA officials involved in the rollout, including Dr. Rebecca Gladding, acting deputy chief of psychiatry at VA Greater Los Angeles, and Dr. Kimberly McManus, the VA’s acting chief AI officer, have pointed to the Kansas City pilot’s patient trust numbers as evidence the technology is improving the actual clinical encounter, not just clerical efficiency. Their argument is that when a doctor isn’t staring at a keyboard, patients report the conversation feels more genuine, and that shows up in survey data as higher trust and satisfaction scores. The near-100 percent opt-in rate, with veterans declining less than 1 percent of the time, is being cited internally as proof that patient acceptance is not the obstacle some skeptics predicted.
The skeptical read
Critics of ambient AI scribes across the broader health system, not specific to the VA, have raised recurring concerns that generalize here: that AI-generated notes can introduce subtle inaccuracies or omissions a rushed clinician may not catch before signing off, that recording sensitive conversations, including behavioral health sessions, raises data privacy and security questions even with consent obtained, and that productivity gains measured in hours saved don’t automatically translate into more time spent with patients rather than more patients scheduled per day. The VA’s own privacy impact assessment for the Abridge deployment was a required step before rollout, an acknowledgment that recording clinical conversations at this scale carries real data-handling risk that has to be documented and mitigated, not assumed away by adoption statistics.
What comes next
The VA’s stated plan is to move from its current footprint toward a broader enterprise contract in 2026 that would standardize ambient scribe access across the department rather than running it site by site through separate vendor pilots, and eventually extend the tools to select outpatient care settings beyond primary care and the specialties already covered. With adoption numbers this large this early, the VA’s experience is likely to become a reference point for other public and private health systems weighing whether ambient AI documentation is ready for enterprise-wide deployment, or whether the real test still lies in what happens to note accuracy and patient trust once the technology reaches millions of visits rather than under a million.
Photo: Tima Miroshnichenko / PEXELS via Pexels
