Ambient AI scribes have earned a place in a lot of emergency departments, and I understand why. Documentation in emergency medicine is relentless, and anything that gives time back at the end of a shift gets attention.
Some of the hype holds up. A single-center study in JMIR AI this year followed more than 10,000 emergency encounters over 13 months and tied scribe use to about 73 seconds less documentation time per patient. That sounds small until you multiply it by 25 or 30 patients. It adds up to real time, or a few more minutes at the dinner table.
But a paper published Oct. 8 in BMJ Innovations should be required reading for anyone signing AI-drafted notes. Researchers tested a commercial ambient scribe inside the U.K. regulator’s AI Airlock sandbox, using more than 5,700 real consultations. With the product’s safety layers turned off, about 3.3 percent of each note’s content had no support in the actual conversation. Most of that, 2.9 percent of each note, was the kind of content that could change a clinical decision. The authors’ examples included suggested diagnoses and hallucinated drug allergies. With the safety checks on, those numbers fell to 1.6 percent and 1.2 percent. Better. Not zero. The authors work for the company that makes the scribe, which makes their candor worth noticing.
Three percent sounds tiny. Now think about where it lands. A made-up penicillin allergy in a septic patient. A line saying the patient takes no blood thinners when nobody asked. A diagnosis the scribe pulled from you thinking out loud, one you ruled out 20 minutes later. In the ED, the next person reading that note might be the admitting team at 4 a.m., a consultant on the phone, or a colleague seeing the same patient next week. They will trust it because my name is on it.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The paper made another point that stuck with me. When regulators and clinicians judged the real product instead of its label, they cared about what it did, not what it claimed to be. A scribe sold as an administrative tool starts acting like decision support the moment it fills gaps with clinical content. The less we read, the more decision support it becomes.
Here is what I tell residents who ask how to use these tools.
Read the dangerous sections first. Allergies, medications, anticoagulation, code status, and the plan. If time is short, those get your eyes before the HPI does.
Delete anything you did not say. If the note has a finding, a history item, or a diagnosis you cannot place in the conversation, take it out. Don’t edit it into something true. Remove it.
Watch for the scribe doing your thinking. Emergency physicians reason out loud in front of patients. We say, “This could be a clot,” all the time. A scribe that turns a differential into a diagnosis has crossed a line.
Give it context. In the BMJ study, notes were more accurate when the clinician added written context. A 10-second spoken summary at the end of the encounter may be the cheapest accuracy check you have.
Ask your hospital what is being monitored. Who reviews scribe errors after go-live? How would anyone know if the error rate climbed after a software update? If nobody can answer, that is the answer.
None of this means giving the time back. I want the 73 seconds. I want doctors leaving on time. But the scribe does not sign the chart. I do. The minute it saves is only a good trade if I spend part of it reading.
Harvey Castro, known as DR GPT™, is an emergency physician, author, keynote speaker, and physician futurist. He is advisor to the CEO of Phantom Space. He writes and speaks about artificial intelligence in health care, digital health, space medicine, and the future of human-centered care. Explore his speaking topics and availability.
He is the author of Bing Copilot and Other LLM: Revolutionizing Healthcare With AI, Solving Infamous Cases with Artificial Intelligence, The AI-Driven Entrepreneur: Unlocking Entrepreneurial Success with Artificial Intelligence Strategies and Insights, ChatGPT and Healthcare: The Key To The New Future of Medicine, ChatGPT and Healthcare: Unlocking The Potential Of Patient Empowerment, Revolutionize Your Health and Fitness with ChatGPT’s Modern Weight Loss Hacks, Success Reinvention, and Apple Vision Healthcare Pioneers: A Community for Professionals & Patients.
He shares updates on LinkedIn, YouTube, Facebook, Instagram, and X. #DRGPT
