Artificial intelligence systems that automatically transcribe medical consultations could overlook critical information and erode clinicians' diagnostic skills, according to research published by academics at the University of Edinburgh.
These ambient AI scribes—software that listens to conversations between doctors and patients, converts speech to text, and generates clinical notes—are currently deployed in approximately 40% of UK general practices. The technology promises to reduce administrative burden, yet the peer-reviewed study, published in the British Medical Journal's digital health and AI publication, identifies significant risks alongside its benefits.
Researchers examined real-world use of the scribes and discovered that while the tools can assist with routine paperwork, they frequently fail to capture non-verbal communication and emotional nuance that inform clinical decision-making. The findings emerge as the NHS continues to expand AI-enabled documentation across primary and secondary care settings.
What information do these systems miss?
A central concern identified by the University of Edinburgh team is that ambient AI scribes prioritise clinical data while systematically undervaluing patients' narratives and lived experiences of illness. The technology struggles to record facial expressions, body language, and emotional states—elements that clinicians rely upon to understand the full context of a patient's condition.
The researchers found that AI-generated summaries tend to emphasise measurable clinical facts at the expense of the patient's own account of their symptoms and circumstances. This selective capture can distort the clinical record and potentially compromise care quality, particularly for patients whose conditions involve psychological or social dimensions.
Additionally, the study documented instances where clinicians failed to recognise their own notes or recall details about patients at subsequent appointments, suggesting that the cognitive process of manual note-taking serves functions beyond mere documentation.
How do patients respond to being recorded?
Research participants revealed that awareness of AI recording and processing discourages patients from disclosing sensitive information. Individuals are significantly less forthcoming about substance use, domestic abuse, mental health difficulties, and other stigmatised conditions when they know the consultation is being captured and analysed by automated systems.
This reluctance to share creates a secondary clinical risk: incomplete patient histories that may lead to missed diagnoses or inappropriate treatment recommendations. The chilling effect on disclosure disproportionately affects vulnerable populations already marginalised within health services.
What are the benefits for clinicians?
The University of Edinburgh report acknowledges genuine advantages. By automating transcription and note generation, ambient AI scribes reduce the time clinicians spend on paperwork, allowing them to redirect attention toward patient interaction and complex clinical reasoning. One clinic in Dudley using the transcription tool Heidi reduced its backlog of patient letters from six months to just 14 days—a substantial operational improvement.
According to NHS England's January 2026 announcement, an NHS-sponsored study across nine sites found that ambient scribing tools increased direct patient interaction time by 23.5% and enabled clinicians to see 13.4% more patients per shift in emergency departments. The same research covered over 17,000 patient encounters across general practices, hospitals, mental health services, and ambulance teams.
However, the University of Edinburgh researchers caution that these efficiency gains come with hidden costs.
What is 'cognitive offloading' and why does it matter?
The study identifies a phenomenon called "cognitive offloading"—the process by which clinicians gradually transfer mental effort to automated systems. While this can initially free capacity for more meaningful conversations, it carries long-term risks to clinical skill and memory.
When clinicians rely on AI to capture and synthesise consultation details, they invest less cognitive effort in reasoning through what patients have told them. Over time, this can reduce memory recall, diminish pattern recognition abilities, and slow the development of diagnostic expertise. The researchers documented cases where clinicians did not recognise notes attributed to their own consultations or failed to remember patients they had seen previously—a concerning indicator of eroded engagement.
How widespread is adoption, and what errors occur?
Adoption of ambient AI scribes is accelerating across the NHS. According to a 2026 UK GP survey of 1,003 practitioners, 14% were already using ambient AI scribes and 39% intended to adopt them soon. Among current users, 86% reported using Heidi Health, the market-leading product.
The same survey revealed a troubling error rate: 32% of users reported that mistakes in AI-generated notes occurred often or always, and 14% said they had encountered errors with significant-to-critical clinical implications. These findings suggest that while the technology reduces administrative time, it introduces new quality assurance challenges that healthcare organisations have not yet fully addressed.
In a separate evaluation published in JMIR Medical Informatics, researchers characterised the ambient scribe market as "immature and volatile," creating procurement uncertainty for health systems considering large-scale deployment.
What do researchers recommend?
The University of Edinburgh team concludes that ambient AI scribes can support administrative efficiency but require substantial further development before widespread rollout. They emphasise that system design must preserve the patient's voice and narrative rather than replace it with algorithmic summaries.
Dr Lucas Seuren, of the University of Edinburgh's centre for biomedicine, self and society, stated:
"Many clinicians are excited about ambient AI scribes, because they promise to cut down on paperwork. 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 researchers call for additional investigation into medium and long-term effects of these tools in specific healthcare settings and warn that risks may differ across countries with different healthcare systems from those where the AI tools were originally developed and tested.
What happens next?
NHS England published updated guidance on AI-enabled ambient scribing products on 29 July 2026, signalling that the technology has moved into formal NHS implementation frameworks. A regional NHS procurement announced in July 2026 is set to deploy ambient voice technology across Midlands clinicians and trusts, indicating that rollout activity will continue following this research publication.
The tension between administrative efficiency and clinical safety remains unresolved. While operational benefits are measurable and significant, the risks to patient disclosure, clinical reasoning, and care quality require ongoing scrutiny as the NHS scales these systems across its workforce.
