A University of Edinburgh review of 27 studies, published in BMJ Digital Health & AI, examines what ambient AI scribes change about the clinical consultation itself. The answer is not about transcription accuracy, and that is the point.
Adoption is already broad
"Some 40 per cent of GPs in the UK have said they used ambient scribes" — a technology adopted at scale in primary care ahead of the evidence base being assembled. That figure is self-reported survey data rather than vendor telemetry, which is worth holding in mind, but the order of magnitude is the story.
What the notes lose
Scribes "miss out on important aspects of consultations such as facial expressions, gestures and emotional state", and "AI summaries tend to prioritise clinical information, which can come at the expense of the patients' stories." The summary is not simply shorter than a human note; it is systematically weighted toward the clinical and away from the narrative.
Patients change what they say
The sharpest finding is behavioural. Patients "may be hesitant to reveal sensitive information, such as substance abuse, domestic abuse or mental health struggles, when they know the consultation is being recorded." No improvement in speech recognition addresses that. The review also raises deskilling: cognitive offloading may "reduce memory recall and skill development", and "clinicians report instances where they do not recognise their own notes or remember the patient the next time they see them."
What the received framing gets wrong
Headlines read "AI notes miss vital information", which sounds like an accuracy failure that better models would fix. This is not a measurement study and reports no error rate at all. It is a review of implementation risk, and its central claims are about behaviour — what patients withhold, and what clinicians stop retaining. Two limits belong with it: this is a synthesis of 27 papers rather than a trial, and the deskilling material is drawn from the reviewed literature rather than from new measurement by the authors.