When AI Scribes Are Writing the Medical Record

AI scribes are creating medical documentationThe clinic note looked flawless. You’ve seen these: there is a complete history, a tidy review of systems, a detailed lung exam, and a clear plan. Then you check the encounter type. It was a telephone visit. Nobody listened to those lungs.

That kind of mismatch is becoming a signature of a new author in the medical record: the ambient AI scribe. If you review records for attorneys, you are already reading chart entries drafted by software, whether the page tells you so.

Here is what the research shows about these tools and what they mean for your LNC review.

What an AI Scribe Does

An ambient AI scribe listens to the conversation between a clinician and a patient through a phone, tablet, or room microphone. It turns that audio into a transcript and then into a structured draft note. The clinician reviews the draft, edits it (or not), and signs it into the electronic health record.

Epic, Microsoft (Dragon Copilot), Abridge, and Ambience Healthcare are among the most common names you will see which are generating these notes.

The key word is draft. The AI is not transcribing word for word. It decides what matters, what to leave out, and how to organize the encounter. That is where the errors come from. On what basis does it decide, you might ask?

Does Anyone Review the Output for Errors?

In theory, yes. The clinician who signs the note is the reviewer. In practice, nothing forces that review to happen.

At University of California San Francisco, no technical mechanism confirms that a physician read a note before signing it. Dr. Robert Wachter put it candidly: after 49 perfect notes, would he really be fully attentive on note 50?

What slips through?

  • Real-world pilot. In a University of California Davis pilot of more than 7,500 AI-drafted clinic notes, physicians graded a sample of 356. Omissions appeared in 18%, hallucinations (content that never happened) in 11.5%, and accidental inclusions in 9.3%. About 5% of notes held errors rated as posing serious risk or imminent harm. Almost 15% of notes were signed with no edits at all.
  • Independent audit. A 2026 preprint that ran 142 consultations through three commercial scribes found verified errors in roughly one note in three. They centered around allergies, medications, invented patient details, and exam findings written up for telephone visits. One error type was new: a treatment the clinician took back during the visit, recorded as care that was delivered.
  • Omissions dominate. A study of seven primary care products found omissions made up about 84% of errors. Hallucinations were rarer but more often clinically serious. No product produced an error-free note.
  • Quality versus humans. In the Annals of Internal Medicine citation, graders scored notes from 11 AI scribe tools lower than notes written by human clinicians in all 10 quality domains

Researchers call the underlying problem automation bias: the tendency to accept machine output as accurate and complete because it usually is. The note reads smoothly, so the wrong medication dose reads smoothly too.

Remember what the scribe cannot capture. It hears words. It does not see the toddler lying limp on a parent’s lap or the patient who goes quiet when asked about if she feels safe at home. If the clinician does not add those observations, they are not in the record.

Clinicians themselves are not fully convinced. In a Brown University study of emergency physicians, fewer than half trusted the accuracy of AI-generated notes, compared with 75% who trusted human scribes.

How Often Is It Being Used?

Despite clinicians’ reservations, adoption of AI scribes has moved faster than almost any technology in healthcare history.

  • Physicians overall. The AMA’s 2026 survey found 81% of physicians use some form of AI in practice, up from 38% in 2023. Ambient scribes are a major driver.
  • Scribes specifically. A January 2026 Doximity survey of more than 3,100 physicians found 29% use voice-based documentation tools such as AI scribes.
  • A national study found about 63% of US hospitals using Epic had adopted or were implementing an ambient AI tool by mid-2025.
  • Scale at single systems. Kaiser Permanente logged more than 2.5 million scribe-assisted encounters in 14 months. Stanford has passed 1 million. Cleveland Clinic deployed its scribe to more than 4,000 clinicians.

Keep one nuance in mind. A 2025 survey of 43 large health systems found every one had at least piloted ambient documentation, but only 14% had fully deployed it. Use varies by clinician, specialty, and site, so you cannot assume every note from a given facility was AI-drafted. You also cannot assume none were.

Patients often don’t know. In a large Canadian survey, only 28% of adults knew these tools were in use.

Who Is Using It? Outpatient, Inpatient, and Now Nurses

Most use so far has been in outpatient physician practices. The largest multisite study of AI scribes, published in JAMA in 2026, focused on more than 8,500 ambulatory clinicians. That is where you are most likely to find AI-drafted notes today: office visits, specialty clinics, and telehealth. 

The tools are moving into hospitals. Houston Methodist has piloted AI scribes in its emergency departments and inpatient units. Expect to see AI-drafted ED notes, admission histories, and progress notes in the records you review.

The newest development affects our own profession: nursing documentation. In 2026, several East Coast health systems began rolling out ambient AI to bedside nurses.

  • Mount Sinai Medical Center in Miami Beach extended Epic’s ambient tool, called Chart with Art, to inpatient nurses in July 2026.  It was the fourth US health system to do so, following Houston Methodist’s bedside nursing launch in March 2026.
  • Penn Medicine has rolled out ambient AI to about 550 inpatient nurses and 80 ambulatory nurses.
  • Jefferson Health is piloting a tool on three units at Abington Hospital that converts spoken language into structured flowsheet documentation.
  • Tampa General is expanding Microsoft’s Dragon Copilot to nurses across its system, and Abridge’s nursing product is live at Mayo Clinic, Johns Hopkins, and others.

Why does this matter for record review? Physician scribes draft a narrative note.

But nursing documentation is more complex. We document in dozens of discrete flowsheet fields per shift: vital signs, assessments, pain scores, fall risk, skin checks, intake and output and many other flowsheets. When a nurse speaks at the bedside and the AI populates those fields, a flowsheet entry that looks like a nurse’s direct observation may be the AI’s interpretation of a conversation.

We are early in the use of AI for nursing documentation. There is almost no published data yet on how accurate AI-populated flowsheets are.

Responsibility for accuracy stays with the nurse. Nurses remain responsible for reviewing, verifying, and co-signing what the AI drafts, just as physicians do.

What This Means for Medical-Legal Review

Here is where your expertise becomes even more valuable to attorneys.

The healthcare professional who signs the note attests to the accuracy of the note. In nearly every scenario, the clinician who signs an AI-generated note carries the liability, not the vendor.

Vendor contracts typically cap the company’s responsibility for documentation errors. A signature is an attestation that the clinician reviewed the note and vouches for it, and a one-click “reviewed” button is increasingly viewed as weak evidence that a real review took place. But the legal system holds that provider to its accuracy.

The signed note may not be the only record. One encounter can produce raw audio, a verbatim transcript, an AI draft, suggested billing codes, and the edited final note.

A malpractice insurer, ProAssurance, warns that retained audio will be discoverable, and that the worst case is a recording that contradicts the chart.

Some vendors let facilities delete audio within a day. Others keep it longer. Attorneys disagree on what should be kept. That uncertainty is exactly why the questions on retention of audio need to be asked early in a case.

Know the red flags. As you review, watch for:

  • Exam findings that could not have been performed, such as a heart and lung exam on a telephone visit
  • Medication lists or allergies that conflict with other parts of the record
  • Treatments recorded as given that the patient says were discussed and then canceled
  • Unusually long, polished notes that read nearly identically across visits
  • No mention of nonverbal observations a clinician would normally document
  • Flowsheet entries that do not match the nurse’s narrative notes or the timeline

Errors travel forward. An AI-generated mistake in one note can be copied into later notes and relied on by the next clinician. Tracing an entry back to its origin may show where a fabricated history or omitted medication first appeared.

The errors compound when AI medical record software cheerfully summarizes inaccurate information, and possibly adding in its own misinformation.

At this point you may be wishing for the good old days of handwritten records.

Consent to have AI scribes in the medical information part of care is now a litigation issue. In November 2025, a patient filed a proposed class action against Sharp HealthCare in San Diego, alleging his visit was recorded by an AI scribe without his consent. He learned about it by reading his own chart, which allegedly contained consent language he never gave.

Similar suits have been filed against Sutter Health and MemorialCare. Texas Senate Bill 1188 now requires practitioners who use AI to review AI-generated records and disclose AI use to patients. A consent statement in the note is something to verify, not accept at face value.

In my personal experience, when my husband sees his gerontologist at Johns Hopkins in Baltimore, at every visit the physician informs him that they are using an AI scribe.

Billing exposure follows. Federal regulators have named AI-influenced documentation and coding as a False Claims Act enforcement priority. Generic, templated language that supports a higher-level code deserves a second look.

Questions to Raise with the Attorney

  • Did this facility or clinician use an ambient AI scribe for the encounters at issue, and which product?
  • Does the audio, transcript, or original AI draft still exist? What is the retention policy?
  • Can the audit trail reveal the AI draft compared with the signed version, and how long the clinician spent reviewing it?
  • Was patient consent for the use of AI scribes obtained and documented, and does state law require it?
  • For nursing care, were flowsheet entries populated by an ambient tool?

A preservation request sent early may capture data that is set to be automatically deleted.

The Takeaway

AI scribes are not going away. They reduce documentation time, and clinicians like them, saying they can focus more on the patient when they do not have to document. But a polished note is not the same as an accurate one. Your clinical judgment, your eye for what does not fit, and your knowledge of how care is actually delivered are what catch the errors a busy clinician missed.

Learn to recognize AI-drafted entries now, because you will be seeing far more of them in every case that crosses your desk.

Keep building the skills attorneys rely on

The 15th LNC Success Online Conference runs March 11–13, 2027. You’ll spend three days learning from experienced legal nurse consultants about building a business attorneys trust and the services that keep them calling. Join us. Register here.

Pat IyerPat Iyer MSN RN LNCC is president of The Pat Iyer Group. She has almost forty years of experience developing resources to assist LNCs in obtaining more clients, making more money, and achieving their business goals and dreams.

Have you heard the most recent podcasts on Legal Nurse Podcast? The show is in its 11th year, putting it in the top 1% of all podcasts for its longevity. Watch our podcast on YouTube at http://LNC.tips/YouTube.

Join our Facebook group, LNC Business Growth Circle, to be part of our LNC community.

Pat’s related websites include the podcasts broadcast at podcast.legalnursebusiness.com, and writing tips supplied at patiyer.com. Pat’s latest books are AI-Powered Video for LNCs: A Guide to Visibility and Growth book and workbook. Order at this link.

Join Pat on Substack for tips on writing books to expand your authority and visibility. Subscribe here. 

Get all of Pat’s content in one place by downloading the mobile app, Expert Edu at www.legalnursebusiness.com/expertedu. Watch videos, listen to podcasts, read blogs, watch online courses and training, and more.