AI Scribes Used by 40% of UK GPs Are Making Critical Errors in Patient Records

AI Scribes Introduce New Risks to Patient Safety

Ambient AI scribes are rapidly transforming healthcare documentation, but emerging evidence reveals significant patient safety concerns. About 40% of GPs in the UK now use these AI tools in healthcare that integrate speech-to-text technology with AI to transcribe patient-clinician consultations1

1

2

2

. However, Healthwatch England has identified multiple instances where AI scribes in NHS records have produced errors in medical records that threaten patient safety3

3

4

4

.

In one alarming case, a patient was incorrectly told she had demyelination—serious nerve damage associated with multiple sclerosis—when her MRI scan actually showed "null demyelination." The AI scribe had dropped the crucial word that reversed the diagnosis entirely. The patient, herself an NHS health professional, described the experience as "very traumatising" and only discovered the error when she questioned the result4

4

.

Critical Errors in Drug Names and Treatment Instructions

The risks of ambient AI scribes extend beyond misdiagnoses to incorrect drug names and missing treatment instructions. In one instance, an AI scribe confused the prescribed medication with a different drug that had a similar name—the exact type of error that pharmacology deliberately works to prevent3

3

. Another summary failed to record a consultant's instruction for a patient to obtain a repeat prescription for migraine medication, potentially leaving them without necessary treatment4

4

.

Dr. Shier Ziser Dawood, a London GP, reported that an AI scribe claimed she had told a patient to "continue their Prozac" despite never prescribing or discussing that medication during the consultation. This represents what experts call "hallucinations"—instances where healthcare AI references information that was never discussed4

4

.

Patients Bear the Burden of Quality Control

Healthwatch England warns that patients themselves are often the ones catching these errors, not clinicians. "These inaccuracies may persist in their records if the patient doesn't catch them," the watchdog stated1

1

. This places the responsibility for accuracy on those least equipped to verify medical documentation. Twenty-seven different AI scribes are currently in use across England's health service, yet there is no England-wide regulatory oversight to ensure they are safe and effective3

3

.

The Medicines and Healthcare products Regulatory Agency has not classified AI scribes as medical devices, leaving them outside the testing regime that would verify their safety before deployment3

3

. Rachel Power, chief executive of the Patients Association, emphasized that "trust and confidence in this technology depend on good communication and genuine partnership with patients and right now both are missing"4

4

.

Beyond Transcription Errors: Missing the Human Element

Researchers from the University of Edinburgh conducted a comprehensive review of 27 articles published in BMJ Digital Health and AI, examining the broader implications of scaling up AI tools in healthcare1

1

. Their findings reveal that while AI scribes can capture spoken words, they frequently miss crucial non-verbal cues including facial expressions, gestures, and emotional state during doctor-patient consultations1

1

2

2

.

The research found that AI summaries prioritize clinical information at the expense of patient narratives and their lived experiences of illness. Patients may also hesitate to disclose sensitive information about substance abuse, domestic abuse, or mental health struggles when they know consultations are being recorded and processed by AI1

1

2

2

. This could further disadvantage already marginalized populations in healthcare services.

The Deskilling Dilemma and Cognitive Offloading

The University of Edinburgh study highlights concerns about deskilling among clinicians who rely on ambient AI scribes. Manual notetaking serves important functions in clinical reasoning and reflection, but using AI scribes can lead to "cognitive offloading"—outsourcing mental effort to the technology1

1

2

2

.

While this administrative burden reduction can free clinicians to focus on complex tasks and meaningful patient conversations, it also carries risks. Clinicians report instances where they don't recognize their own notes or remember patients during follow-up visits1

1

2

2

. This raises questions about whether the time saved in documentation comes at the cost of reduced memory recall and skill development.

Dr. Lucas Seuren, GAIL Fellow at the University of Edinburgh's Centre for Biomedicine, Self and Society, noted: "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"1

1

2

2

.

What Comes Next for Healthcare AI

The government's 10-year health plan expects AI scribes to liberate NHS staff from administrative tasks, positioning them as central to the shift from analog to digital healthcare4

4

. Yet the current situation presents a regulatory gap: tools marketed as passive transcribers avoid the medical devices classification that would require safety testing, while those suggesting diagnoses or treatments may face scrutiny3

3

.

Research by Dr. Charlotte Blease at Uppsala University found that errors are more likely with multi-person consultations, patients with complex medical histories, and those whose first language isn't English4

4

. However, more than half of 1,003 UK GPs surveyed believed their ambient AI records were more accurate than their own manual notes4

4

.

Healthwatch recommends that patients be informed when AI scribes are used and given their notes to review—transforming an accidental safety mechanism into a deliberate one. The researchers emphasize that designing these systems must preserve patient trust and the patient's voice rather than overwrite it. More work is needed to understand medium and long-term impacts, particularly for tools used in countries with different healthcare systems from where they were developed1

1

2

2

.