Ambient AI scribes became one of the fastest-adopted tools in outpatient medicine over the past two years, quietly listening to visits and drafting notes so doctors could look at patients instead of screens. A study published in the September 14, 2026 issue of the Journal of Hospital Medicine asked a harder question: does the same technology work once it moves into the far messier environment of a hospital ward? The early answer is a qualified, complicated no.
Inside the Sutter Health rollout
Researchers examined a rollout of Abridge, an AI scribe that records clinician-patient encounters and generates a transcript along with draft progress-note components clinicians can edit before inserting into the electronic health record, across 20 hospitals in Sutter Health’s network beginning in March 2025. Sutter, a large health system spanning northern and central California, was the first U.S. health system to fully integrate Abridge directly within its Epic EHR, a technical achievement that made this real-world hospital test possible in the first place.
The numbers were underwhelming
Only 59 physicians ultimately participated in the study, about half of them hospitalists. Among that group, hospital physicians used the AI scribe for less than half of their patient notes — far below the near-universal adoption ambient scribes have achieved in some outpatient primary-care settings. Just over half, 53 percent, said they had a “high degree of success” using the tool for documentation. The study authors concluded that while ambient AI scribes hold real promise for inpatient medicine, the technology has “significant room for improvement” for physicians working in hospitals.
Why hospital medicine is a different animal
The gap between outpatient success and inpatient struggle comes down to the nature of the encounter. A clinic visit is typically a single, contained conversation between one doctor and one patient. A hospital note synthesizes information scattered across multiple daily encounters, nursing updates, lab results, specialist consults and overnight events — material an ambient microphone sitting in a single conversation simply doesn’t capture. Hospitalists also round on many more patients per shift than outpatient physicians see in a day, and inpatient documentation carries more legal and billing complexity, from admission histories to discharge summaries, that a transcript-and-draft model has to piece together rather than just record. Multiple hospitalists interviewed in coverage of the study described a workflow mismatch: they might see a patient briefly on morning rounds, then need to fold in an afternoon lab result, a family conversation that happened without the microphone present, and an overnight nursing observation before the note is truly complete. An ambient tool built to transcribe one continuous conversation has no mechanism for capturing any of that.
The broader ambient-AI boom keeps growing regardless
The Sutter finding lands amid otherwise blistering growth for ambient documentation tools. Ambient AI became generally available for nurses across more than 250 U.S. health systems earlier this year, with rollouts extending into inpatient nursing units, emergency departments and post-acute settings through the summer. A separate JAMA study across five academic medical centers found ambient scribes cut total EHR time by 13.4 minutes and documentation time by 16 minutes per clinician, translating into more than 15,700 hours of saved documentation time over a year compared with non-users — the equivalent of nearly 1,800 working days. At UCSF, roughly 70 percent of physicians now use an AI scribe in daily practice. That gap between outpatient enthusiasm and inpatient hesitancy is exactly what makes the Sutter data notable: it’s one of the first large, published looks at the technology specifically in hospital medicine, rather than in the primary-care clinics where most of the glowing adoption numbers originate.
Two views of the same data
Vendors and hospital administrators point to the outpatient numbers and the sheer speed of adoption as proof the technology is transformative and simply needs more tuning for inpatient workflows. Frontline hospitalists and documentation researchers read the Sutter results more skeptically: if only half of participating physicians report strong success and most still write more than half their own notes by hand, the tool is, for now, a partial assistant rather than the workflow overhaul vendors are selling to hospital boards weighing six- and seven-figure contracts.
What comes next
Expect AI scribe vendors to push updates specifically built for multi-encounter inpatient documentation — tools that pull structured data from nursing notes, vitals and prior encounters rather than relying solely on a recorded conversation. Sutter’s experience is likely to become a reference point other health systems cite before expanding scribe contracts into their own inpatient units, and future studies will need larger physician cohorts than 59 to say definitively whether the technology’s inpatient stumbles are a temporary rollout problem or a structural mismatch between how hospital medicine works and how these tools were designed. Sutter has not said whether it plans to expand or pause the hospital rollout, but given the system’s role as the first in the country to fully integrate Abridge within Epic, whatever it decides next will likely be read by other large health systems as an early signal of where inpatient ambient AI is actually headed, separate from the outpatient hype that has driven most of the sector’s growth so far.
Photo: Kost9n4 / PIXABAY via Pixabay
