AI Scribes Reach 986K VA Visits

Nearly a million visits, two different scribes

The VA started small. In October 2025, ambient AI scribes went live at 10 medical centers. Doctors talked, patients talked, and software drafted the note in the background. No typing during the visit. That was the pitch.

A year later, the numbers are in. VA appointments using ambient scribes have crossed 986,000 since that October start. The tools are now open to primary care providers, behavioral health interdisciplinary teams, physical medicine and rehab practitioners, and medical and surgical specialists across VA Patient Aligned Care Teams. Veterans declined the recording less than 1% of the time.

That is a real number, and it says something true: patients mostly don't mind being recorded when it means their doctor is looking at them instead of a screen. A Kansas City pilot backs that up. Veteran satisfaction rose to 95.8%, up about 3 points. Trust in the care team hit 95%, up nearly 3 points. Providers were rated clearer in their explanations 95.4% of the time, up 2.3 points.

But it's not one tool. It's two.

Here's the part that matters more than the adoption curve. The VA isn't running a single scribe. It's running Abridge, live at 70 VA sites and used across more than 300 U.S. health systems overall, and Knowtex, deployed in 10 VISNs and 79 VA medical centers with more than 7,000 clinician users. Knowtex alone reports 450,000-plus documentation hours saved over six months, 88% adoption among users who try it, and a 4.48-out-of-5 satisfaction score. Both vendors are live, in different parts of the same health system, and the VA says an enterprise-wide contract is still coming in 2026.

Think about what that means for the doctor in the middle. Two products, two workflows, two places the transcript might end up, and a supervising clinician who has to know which one is running in which building. Every one of those numbers, the 986,000 visits, the 450,000 hours, the 88% adoption, is a real win for documentation time. None of them tell you whether the note the scribe wrote actually reconciles with the chart, the orders, or what got handed off at the end of the shift.

The scribe problem was never just the scribe

This is the pattern across ambient AI in medicine right now. The scribe layer is maturing fast, because it's the easiest problem: listen, transcribe, draft. It gets deployed first because it's the most fundable, most demoable, least risky piece of the doctor's day. But a doctor's day is not one task. It's a chart to write, an order to place, a consult to check, a colleague to hand off to at 8pm. A scribe that saves nine minutes per note is a real gain. It is also, on its own, a point tool bolted onto a workflow that was never redesigned around it.

The VA's own numbers make the case better than any critique could. Two vendors, both succeeding on their own terms, running in parallel inside the same hospital system, each with its own adoption metrics and its own satisfaction score. That's not a workspace. That's two good products next to each other.

None of this is a knock on Abridge or Knowtex, both of which are clearly doing what they were built to do. It's a question for anyone scaling AI into a hospital: are you buying tools one function at a time, or building the place where a doctor's whole shift actually lives? The VA's rollout, done carefully and at real scale, is still answering the first question. The second one is still open, at the VA and everywhere else.

Sources

• VA to gradually expand medical AI scribe following successful pilots (Nextgov/FCW, Sept. 14, 2026)