AI Scribes Reduce Doctor Paperwork but Trigger Accuracy and Equity Concerns in Healthcare

AI Scribes Promise Relief from Pajama Time

Hospitals across the United States are rapidly deploying AI scribes to address a persistent problem: pajama time, the hours doctors spend compiling patient notes after their shifts end1

1

2

2

. These AI-powered clinical note-taking tools listen during patient encounters and automatically generate clinical notes, promising to free physicians from administrative burden. As healthcare AI adoption accelerates, the data these systems produce has become the foundation for billing, research, and clinical decision-making across health systems.

But this efficiency comes with significant risks that could fundamentally alter healthcare infrastructure. Taofeeq Oluwatosin Togunwa, M.D., and Jodyn Platt, Ph.D., from the Department of Learning Health Sciences at University of Michigan Medical School, published research in Learning Health Systems examining how automating patient note-taking changes the nature of medical data itself1

1

2

2

.

Machine-Generated Clinical Notes Introduce New Data Gatekeepers

"It used to be that the doctor or clinician taking the notes was the filter for data. Now a machine is in the pipeline deciding what does and does not become data," Platt explained1

1

2

2

. This shift represents a fundamental change in how medical information enters the healthcare system. Where human clinicians once controlled what became part of the official record, AI in healthcare now makes those determinations—often through proprietary training data that remains opaque to users and regulators.

The researchers identified AI scribe accuracy concerns centered on demographic bias. These models typically train on English and American accents, performing well within that narrow distribution. "But you find concerning literature where these models actually don't perform as well with people that might have different types of accents," Togunwa noted1

1

2

2

. When AI encounters patients outside its training parameters, the results can be problematic.

Hallucinations and Performance Drift Threaten Data Integrity

When inputs don't match training data, AI scribes produce hallucinations—fabrications that then become part of official medical records1

1

2

2

. These errors don't remain isolated mistakes. They cascade through healthcare systems, potentially affecting billing accuracy, research validity, and clinical decision-making for future patient care. The implications extend far beyond a single encounter.

While clinicians are expected to review machine-generated notes for accuracy, practical realities undermine this safeguard. "There isn't a lot of incentive to go line by line if the note looks generally fine," Togunwa observed1

1

2

2

. Doctors already stretched thin are unlikely to scrutinize AI-generated text that appears superficially correct, allowing subtle errors to slip through.

The accuracy problems compound over time through performance drift. "If real-life data is substantially different from the training data, you start to have these challenges where there's a drift in performance of the model," Togunwa explained1

1

2

2

. Without routine updates and continuous auditing, AI scribes become progressively less reliable as medical terminology evolves and patient populations shift.

Equitable Benefits Remain Uncertain

The researchers emphasize that doctor paperwork reduction alone doesn't justify widespread AI scribe deployment. Platt advocates for comprehensive measurement: "We need to measure more things more often. And not the same old things and not just the things that are easy to measure, but the things that we know are important"1

1

2

2

.

Health systems must ensure equitable benefits reach patients, not just physicians. "Are our patients also receiving that same level of benefit in their health care? For example, are they able to get to a diagnosis faster, or to see a clinician faster and are those benefits accrued equitably?" Platt questioned1

1

2

2

. Without deliberate attention to equity, AI scribes risk widening existing healthcare disparities, particularly for patients with non-standard accents or those from underrepresented demographics in training datasets.

The research suggests healthcare organizations should implement rigorous monitoring frameworks before scaling these technologies. As AI scribes become embedded infrastructure rather than experimental tools, the stakes for getting implementation right increase dramatically. The question isn't whether AI can reduce administrative burden—it clearly can—but whether health systems will invest in the continuous auditing and measurement necessary to ensure these tools serve all patients equitably while maintaining data integrity.