Across the United Kingdom, a quiet transformation is underway in the examination room: artificial intelligence now listens as doctors and patients speak, converting human encounter into clinical record. A review of 27 studies by researchers at the University of Edinburgh reveals that what these systems capture — words — is only a fraction of what medicine requires, and that the gap between transcript and truth may fall heaviest on those already least heard. In the long arc of healthcare's relationship with technology, this moment asks an old question anew: when we automate the act of listening
AI Clinical Scribes Risk Missing Vital Patient Information, Study Warns
Vulnerable and marginalized patients may face further disadvantage if AI scribes fail to capture their experiences and sensitive health information.
The AI hears the words but not the person.
Mark
So the core finding is that AI scribes miss things that matter. What kinds of things?
Mimi
Non-verbal cues, mainly. Facial expressions, gestures, emotional state. The AI hears the words but not the person.
Luke
But how much does that actually affect clinical outcomes? The study reviewed 27 papers—did any of them measure whether patients got worse care?
Mimi
That's the gap. The review identifies the risk, but you're right that direct evidence of harm is limited. It's more about what could be lost.
Mark
And patients withhold information when they know they're being recorded?
Mimi
Yes. Sensitive stuff especially—abuse, addiction, mental health. The recording itself changes behavior.
Luke
Is that specific to AI, or would patients do the same if a human scribe was in the room?
Mimi
Good question. The research suggests it's the AI specifically—the sense that a machine is processing your words. But the studies don't isolate that perfectly.
Mark
What about the clinicians? You said some couldn't remember their own notes.
Mimi
Right. When you stop writing by hand, you stop doing the cognitive work of synthesis. Some doctors reported not recognizing their own summaries later.
Luke
But the study also says this frees up time for better conversations. So there's a real trade-off.
Mimi
Exactly. More face-to-face time, less administrative burden. But potentially less deep clinical reasoning and memory.
Mark
Who gets hurt most by this?
Mimi
Vulnerable patients. People already struggling to be heard in healthcare. If the AI misses their story, they lose another chance to be understood.
Luke
And we don't know yet how this plays out in different healthcare systems or countries.
Mimi
No. Most research has been done where the tools were built. That's a real blind spot.
The Pulse
- AI scribes are now used by roughly four in ten UK general practitioners, deployed at speed to solve a genuine crisis of administrative burden — but the science of their consequences is only beginning to catch up.
- The technology transcribes words faithfully yet remains blind to the tensed shoulder, the withheld tear, and the silence that often carries the most clinical weight.
- Patients who know a machine is processing their words grow guarded, withholding disclosures of abuse, addiction, or mental illness — making the clinical picture incomplete by the very design of the encounter.
- Clinicians report a creeping cognitive offloading: some cannot recognize their own AI-generated notes, others fail to remember patients at follow-up, as the reflective friction of active note-taking quietly disappears.
- The University of Edinburgh team calls for fundamental redesign — systems that preserve patient voice rather than flatten it — and for rigorous testing across diverse healthcare settings before further scale.
Across the United Kingdom, a quiet transformation is underway in the examination room: artificial intelligence now listens as doctors and patients speak, converting human encounter into clinical record. A review of 27 studies by researchers at the University of Edinburgh reveals that what these systems capture — words — is only a fraction of what medicine requires, and that the gap between transcript and truth may fall heaviest on those already least heard. In the long arc of healthcare's relationship with technology, this moment asks an old question anew: when we automate the act of listening, what do we lose of the act of caring?
Across the United Kingdom, roughly four in ten general practitioners have turned to ambient AI scribes — tools that listen during consultations and convert speech into clinical notes. The appeal is understandable: documentation burdens are crushing, and anything that restores time for actual medicine feels like relief. But a University of Edinburgh review of 27 studies suggests the trade-off is steeper than most clinicians have reckoned with, and that its costs land hardest on patients already struggling to be heard.
The core finding is deceptively simple. These systems are good at one thing: capturing words. What they cannot capture is everything else — the facial expression that contradicts a patient's reassurance, the gesture that signals distress, the emotional texture that experienced clinicians read as fluently as any symptom. A consultation is not a transcript, and treating it as one means the record that emerges is already incomplete.
The problem deepens when patients know they are being recorded by a machine. Research found that awareness of AI processing makes people more guarded — less willing to disclose domestic violence, substance use, or mental health struggles. The technology, simply by being present, changes what patients are willing to say. And when AI summaries are generated, they tend to prioritize clinical facts over the patient's own account of their illness — flattening the human story into extractable data.
For clinicians, the picture is double-edged. Time saved on documentation is real and valuable. But researchers also documented cognitive offloading: doctors who stop taking their own notes stop engaging in the active synthesis that sharpens clinical reasoning. Some reported not recognizing their AI-generated notes; others could not recall patients at follow-up visits. The skill, it seems, depends on the friction.
The Edinburgh team calls for these tools to be fundamentally redesigned — built to preserve patient voice, tested across varied healthcare systems, and deployed only with a clearer understanding of who bears the cost when they fall short. As one researcher noted, clinicians are solving a real problem, but the patient experience has been poorly considered. Efficiency, pursued without that consideration, risks becoming its own form of harm.
Across the United Kingdom, roughly four in ten general practitioners have adopted artificial intelligence tools designed to listen during patient consultations and convert speech into clinical notes. These systems, called ambient AI scribes, promise to free doctors from the burden of paperwork—a genuine relief in a profession drowning in administrative work. But a new review of the scientific literature suggests the trade-off may be steeper than clinicians realize, and the cost may fall hardest on patients who are already vulnerable.
Researchers at the University of Edinburgh examined 27 published studies examining how these AI scribes function in practice and what happens when they scale. What they found was a gap between what the technology captures and what actually matters in a clinical encounter. The AI systems excel at one thing: transcribing words. They record what the patient says and what the doctor says with reasonable fidelity. But a consultation is not a transcript. It is a conversation shaped by silences, by the way someone's shoulders tense when a difficult topic surfaces, by the tears that don't make it into any note. The AI misses facial expressions. It misses gestures. It misses the emotional texture that often tells a clinician more than the words alone.
There is a second problem, subtler and perhaps more consequential. When the AI does produce a summary, it tends to prioritize clinical facts—the diagnosis, the symptoms, the treatment plan—at the expense of the patient's own story. A person's experience of their illness, the way it has disrupted their life, the fears that keep them awake, these details get flattened or omitted entirely. The system is designed to extract data, not to preserve voice. For patients already accustomed to being unheard in healthcare settings, this represents another small erasure.
The research also uncovered a behavioral shift. Patients who know they are being recorded and that an AI system will process their words become more guarded. They are less likely to disclose sensitive information—substance abuse, domestic violence, mental health struggles—when they understand that their words are being captured by a machine. The very act of introducing the technology changes what patients are willing to say, which means the clinical picture becomes incomplete by design.
For clinicians, the consequences cut in two directions. On one hand, the AI scribes do what they promise: they reduce the time spent on documentation, which creates space for more direct conversation with patients and more focus on complex clinical reasoning. That is real and valuable. On the other hand, the researchers found evidence of what they call cognitive offloading—the outsourcing of mental work to the machine. When clinicians stop taking their own notes, they stop engaging in the active process of listening and synthesizing. Some doctors reported that they did not recognize their own AI-generated notes when they read them later. Others found they could not remember the patient at the next visit. The skill of clinical reasoning, which depends on the friction of active note-taking and reflection, begins to atrophy.
The University of Edinburgh team concludes that these systems need fundamental redesign. They must be built to preserve the patient's voice rather than replace it. They must capture not just words but the fuller texture of the encounter. And they must be tested rigorously in different healthcare systems and settings before they are deployed at scale. Right now, much of the research on ambient AI scribes has been conducted in the countries where the tools were developed. How they will function in different healthcare contexts, with different patient populations and different clinical cultures, remains largely unknown.
Dr. Lucas Seuren, one of the researchers, put it plainly: clinicians are excited about these tools because they solve a real problem. But the patient experience has been poorly considered, and there is a genuine risk that the stories of patients—especially those already marginalized in healthcare—will be lost entirely. The promise of efficiency, in other words, may come at the cost of care.
Notable Quotes
Clinicians are excited about these tools because they solve a real problem. But the patient experience has been poorly considered, and there is a genuine risk that the stories of patients—especially those already marginalized in healthcare—will be lost entirely.— Dr. Lucas Seuren, University of Edinburgh Centre for Biomedicine, Self and Society
Want the full story?
Read the original at News-Medical ↗
Technology