The case for AI-powered documentation tools has moved from theory to data. By mid-2025, 62.6 percent of U.S. hospitals running Epic had deployed some form of ambient AI scribing, according to an Emory University study of 2,784 health systems. The ambient documentation market crossed $600 million in revenue that same year, more than double what it produced in 2024. Clinicians who adopted these tools reported real reductions in documentation time and, in several studies, meaningful decreases in burnout.
But hospitals with the highest operating margins and patient volumes were also the ones most likely to have adopted early. Smaller, lower-resourced systems are still catching up, often without the legal and IT infrastructure that larger institutions bring to implementation. That gap matters because the risks of a poorly structured rollout do not scale with institution size.
Over more than two decades of working with health systems on clinical technology deployment, I have watched the same sequence repeat: a compelling vendor demo, an accelerated procurement decision, and a rollout that hits problems no one anticipated when the contract was signed. These tools are worth implementing. They are not worth implementing without the right preparation.
The accuracy problem is real and manageable
Modern AI scribes report hallucination rates in the range of one to three percent per note. That rate matters more at volume. A physician seeing 20 patients a day at a one-percent error rate could encounter fabricated content in notes several times a week.
Physical exam sections carry the highest documented risk. Research comparing ambient-generated notes to manually written ones found hallucinations present in 31 percent of ambient notes versus 20 percent of handwritten ones. These errors are not always obvious. A reflex finding documented without examination. A symptom noted that was never discussed. The kind of inaccuracy that a quick approval, rather than an actual read-through, will not catch.
No vendor accepts clinical liability for those errors. Every contract states this explicitly. The clinician who signs the note holds full legal responsibility, regardless of how it was drafted. That does not make these tools inadvisable. It makes thorough physician review mandatory, and your training program needs to reflect that distinction clearly.
HIPAA, BAAs, and consent are not administrative checkboxes
Any AI tool that captures, processes, or stores protected health information requires a signed Business Associate Agreement before first use. That requirement is not negotiable, and it gets skipped in fast-moving procurement more often than it should.
Patient consent rules are more complicated and less uniformly understood. Ambient AI captures everything said in a clinical encounter. Several states require affirmative patient notice before any AI system records or processes that conversation. Some require explicit opt-in consent. Others require a clear opt-out option. The requirements vary by state and by encounter type, and they are still evolving at the legislative level.
The current regulatory classification also carries forward risk. As of 2026, ambient AI scribes are treated as administrative tools, not medical devices, placing them outside FDA oversight. That classification may shift as these systems take on expanded clinical functions. Build a scheduled regulatory review into your governance plan from the start, not as an afterthought.
EHR integration is where most plans break down
Vendors describe their tools as EHR-agnostic. In practice, integration depth varies considerably by platform. Tools that work cleanly in one environment often require substantial custom configuration in another, and that work lands on IT teams that are already stretched.
Before signing any contract, ask for documented integration specifications for your exact EHR version, not a general compatibility statement. Ask which note types are supported, where data writes to, and who manages the connection when your EHR releases a platform update. Get those answers in writing and with a defined support SLA.
It is also worth understanding what your existing EHR vendor is already building. Epic announced more than 150 embedded AI features in its platform. Oracle Health is deploying AI agents across clinical operations and revenue cycle. athenahealth made its ambient scribing capability available to all users at no added cost. For some organizations, waiting 12 months for an EHR-native tool is the lower-risk path compared to configuring a standalone integration today.
The ROI picture is mixed, and adoption is not guaranteed
A 2025 randomized trial published in NEJM AI found that physicians using AI scribes completed notes faster and reported lower burnout scores compared to the control group. A 2024 JAMA Network Open study of 263 clinicians across six health systems found burnout prevalence dropped from 51.9 percent to 38.8 percent after 30 days with ambient documentation support. Those are real outcomes.
The same 2025 NEJM AI trial also found that 15 percent of physicians assigned to an AI scribe never used it. Adoption is not automatic, and non-adoption cancels every projected return on the investment.
Barriers to adoption tend to cluster around two issues. The first is trust: clinicians who have encountered AI-generated errors become reluctant users, even after accuracy improves. The second is workflow disruption: time-saving tools still require adjustment, and physicians under high patient volume have limited tolerance for a learning curve during peak hours.
Piloting with a small group of engaged early adopters, measuring actual note quality alongside documentation time, and using that data to adjust before system-wide rollout is more reliable than a simultaneous full deployment. Cost projections should also account for IT configuration work, staff training time, and the physician review time that partially offsets documentation savings.
Frequently asked questions
Are AI-powered documentation tools regulated by the FDA?
As of 2026, ambient AI scribes are classified as administrative tools, not medical devices, and are not subject to FDA oversight. That classification may change as their clinical functions expand.
Who is liable if an AI-generated clinical note contains an error?
The clinician who signs the note. No ambient AI vendor accepts clinical liability. Physician review before signature is legally and professionally mandatory.
What compliance documents are required before going live?
A signed Business Associate Agreement is required before any tool handling PHI is activated. Patient consent requirements depend on your state and encounter type.
How much documentation time do these tools actually save?
Peer-reviewed studies show reductions ranging from under one minute to roughly two minutes per encounter on average. Results vary by specialty, visit complexity, and how consistently the tool is used.
Should we deploy a standalone AI scribe or wait for our EHR vendor to build one?
That depends on your timeline and existing contract. EHR-native tools offer tighter integration and lower configuration risk, but most major vendors are still 12 to 24 months from full deployment. A standalone tool can deliver value now, provided integration requirements are fully documented upfront and a governance structure is in place.
Photo: z_wei, Getty Images
Dr. Giriraj Tosh Purohit is an experienced Product Manager and Business Analyst with a strong background in healthcare technology and management consulting. With expertise spanning clinical workflows, EHR, RCM, Digital Health, AI products, and agile product development, He has been instrumental in shaping innovative healthcare solutions.Leveraging deep clinical knowledge, Dr. Giriraj collaborates closely with product management teams to translate customer requirements into functional designs, focusing on interoperability, integration, clinical decision support, and data validation. His familiarity with HL7 fundamentals, including FHIR, V2, V3, and CDA standards, complements his strong foundation in compliance, risk management, MIPS reporting, EHNAC certification processes, and user acceptance testing.
This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.
