Ambient AI Scribes Miss Nonverbal Cues, May Erode Clinical Skills

AI Documentation Tools Show Significant Gaps in Patient Care

Ambient AI scribes—tools that combine speech-to-text with artificial intelligence to automatically generate clinical notes—are being adopted rapidly across healthcare settings, with 40 percent of general practitioners in the UK now using them. But a comprehensive review from the University of Edinburgh reveals these systems may be creating new risks even as they reduce administrative burden.

Researchers analyzed 27 scientific and medical articles examining the adoption and implications of AI scribes in clinical practice. Their findings, first reported in BMJ Digital Health and AI, highlight fundamental limitations in how these tools capture the full scope of patient encounters.

What Gets Lost in AI Translation

The most significant finding centers on what ambient scribes fail to capture. While the technology excels at transcribing spoken words, it systematically misses nonverbal communication—facial expressions, body language, and emotional states that often carry diagnostic weight.

The AI-generated summaries also tend to prioritize clinical data points over patient narratives. The tools extract symptoms and measurements efficiently but can strip away the context of how patients experience their illness, potentially losing information that matters for diagnosis and treatment planning.

Patient disclosure represents another vulnerability. The research team found that knowing a consultation is being recorded and processed by AI makes some patients reluctant to share sensitive information about substance abuse, domestic violence, or mental health concerns.

The Cognitive Offloading Problem

Beyond what the technology captures, the review examined how it changes clinician behavior. Manual note-taking serves functions beyond documentation—it's part of the clinical reasoning process, helping physicians process information and reflect on what they've heard.

AI scribes enable what researchers call "cognitive offloading," where mental effort is outsourced to technology. This creates a double-edged outcome: clinicians gain time for more complex tasks and deeper patient conversations, but they also experience reduced memory recall and slower skill development. Some clinicians reported not recognizing their own AI-generated notes or remembering patients at follow-up visits.

Why it matters

With ambient AI scribes spreading rapidly across healthcare systems, understanding their limitations is critical before they become standard practice. The technology promises efficiency gains that could address clinician burnout, but the Edinburgh review suggests current implementations may be optimized for documentation speed rather than care quality. For healthcare organizations evaluating these tools, the findings point to a need for system designs that preserve patient voice and support—rather than replace—clinical reasoning processes.

Research Gaps Remain

The researchers emphasized that more work is needed to understand how these tools perform in specific healthcare settings over time, particularly when deployed in countries with different systems than where the AI was developed. They called for designs that preserve rather than overwrite the patient's voice.

"Many clinicians are excited about ambient AI scribes, because they promise to cut down on paperwork," said Dr. Lucas Seuren, GAIL Fellow and Research Fellow at the University of Edinburgh's Centre for Biomedicine, Self and Society. "But the experiences of patients are poorly considered, and there are real risks that the patients' stories are lost. This can further disadvantage people who already face marginalisation in health and social care services."

The findings were published in BMJ Digital Health and AI by researchers from the University of Edinburgh.