Are AI-Generated Medical Notes Worth it?

Years ago, after finishing a patient visit, doctors would jot down a few quick notes mainly for their own reference. Now, those casual notes have been replaced by detailed electronic health records (EHRs) that coordinate care between specialties and facilities, while also serving purposes like billing and legal documentation.

In primary care today, physicians spend around two hours daily filling in EHRs for patients. Between 2009 and 2018, the average length of these records grew by 60 percent. Yet much of this work is simple recordkeeping rather than clinical reasoning. So why take doctors away from patients—or hire entire teams of scribes to handle it—when AI could now take on this role instead?

AI Scribes Enter the Scene 

New AI-powered scribes are now being tested in clinics across the US and globe, offered by companies such as SOAPsuds AI Medical Scribe Freed, Tali, Revmaxx, and others in a market exceeding $2 billion. It’s not just private providers making use of them; the U.S. Department of Veterans Affairs signed agreements for trial programs with Nuance and Abridge in 2024. Some reports suggest nearly 30 percent of practices already use this kind of technology. Most of these systems work in a similar way: they listen during patient visits, transcribe the discussion, and then format the details into the standard medical note format. Notes can be ready within seconds or minutes of the conversation. For many clinicians, this is both promising and a little unsettling. AI has the ability to streamline work, but it’s also been known to make confident but incorrect statements, draw from flawed data, and miss the latest clinical guidelines.

Human Errors Are Not Uncommon 

Still, these shortcomings—bias, mistakes, or outdated information—are also found in human documentation. In fact, human notes can be far from perfect. A VA study showed that 90 percent of doctor-written notes contained at least one error when compared with recorded visits. Another review found that 96 percent of speech recognition–based notes had mistakes, and even after review, 42 percent still contained inaccuracies. An emergency department study discovered that some exams were documented even when they didn’t take place, with barely over half being confirmed by observation. Patient concerns also often fail to make it into the record at all. When compared with this reality, AI scribes may seem less risky. Errors already exist in records that guide medical decisions, determine risk, and train predictive models. If AI reflects bias, it’s often inherited from the human-written notes it learns from, which already contain those biases.

Data Privacy Remains a Concern 

Privacy risks from AI tools are significant. Audio from medical visits may be stored by outside vendors, creating potential vulnerabilities. But medical data breaches are already frequent and large in scale. Since 2021, over 700 breaches have occurred annually. In 2024 alone, 703 incidents impacted more than half of the U.S. population—over 181 million people. Patients usually have no option to withhold their data from AI use if they want care. HIPAA, the main U.S. health privacy law, has not been updated in years, but AI documentation tools likely won’t worsen an already fragile system.

Is AI More Reliable than Human Scribes? 

AI may be as unreliable as human scribes and other current methods of medical note-taking like remote or telescribes, but it is far quicker and efficent. This faster output could provide notable benefits. Doctors might spend more face-to-face time with patients. After-hours charting could be reduced. Burnout—which rose to record levels during the pandemic—might ease. These outcomes would all be positive. If AI allows doctors to focus more on patient care instead of constant data entry, and its accuracy matches what’s already common, it could represent meaningful progress.