Ambient artificial intelligence scribes are beginning to reshape clinical documentation in primary care, with many users reporting less time spent writing notes and lower mental effort during consultations. A UK survey published in BMJ Health & Care Informatics examined adoption, workflow effects, documentation errors and consent practices among general practitioners. Current use remained limited, although a substantial share of respondents expected to adopt the technology soon. The results point to meaningful efficiency gains but also show that generated notes frequently require correction, particularly in complex, multilingual or multiparty consultations. Uneven consent practices and continued clinician responsibility for final records add further safety and governance concerns.
Early Adoption Is Accompanied by Strong Interest
Ambient AI scribes capture consultation dialogue and automatically produce draft clinical notes for electronic records. In the survey of more than 1,000 general practitioners, around one in seven were already using the tools, while a much larger group expected to begin using them in the near future. Almost half had no plans to adopt them.
Use differed across parts of the workforce. Adoption was lower among older doctors and locums, while partners or principals were more likely to report current use. Users were also less commonly based in the smallest practices. No difference was found by gender. These patterns came from a relatively small group of current users and did not establish the reasons behind the variation.
Must Read: AI Scribes Spread Rapidly Across UK Primary Care
Most users relied on the same commercial platform, with other systems reported far less often. This concentration means that the findings may reflect the characteristics of one dominant product rather than ambient scribing technology as a whole. Performance, integration and usability may differ between systems.
The participating users generally worked in busy clinical settings and most saw more than 25 patients a day. A minority reported significant symptoms of burnout, but the survey did not measure changes over time or show whether the technology affected burnout directly. Recruitment through an online professional network may also have favoured doctors who were more digitally engaged, so the adoption figures describe an early-user snapshot rather than the wider general practice workforce.
Efficiency Gains Depend on Careful Review
Most current users said ambient AI reduced the burden of documentation. Four in five reported spending less time creating notes and seven in ten experienced a lower cognitive workload during consultations. More than half considered the generated documentation better than their usual notes and said that only limited revision was needed.
However, the systems did not remove the need for checking. Errors were common across generated records and many doctors reported correcting a large share of the output. Nearly one-third said errors occurred often or consistently. Although most problems were judged to have little or moderate clinical significance, a smaller group reported errors with potentially significant or critical consequences.
The combination of time savings and frequent correction suggests that users treated the generated text as a draft rather than a finished clinical record. Efficiency may therefore come from editing a structured summary instead of composing notes from the beginning. Clinicians still need to review, amend and approve the final record.
Comments from respondents reflected cautious optimism. Some valued the ability to focus more fully on the patient rather than the keyboard and regarded the technology as useful and time-saving. Others said the notes could miss important details, make consultations appear too similar or fail to capture uncertainty and clinical nuance. Concerns also extended to professional identity and integration with existing systems. The benefits therefore remained closely linked to active oversight, reliable workflows and continued improvement in accuracy.
Complex Consultations Carry Greater Risk
Errors were more frequently reported in consultations involving several patients or carers, complex medical histories or patients whose main language was not English. Difficulties also arose in noisy environments and in encounters involving speech or cognitive impairment. The technology may therefore be less reliable when communication is already demanding.
Multispeaker dialogue and complicated histories can increase the risk of omissions, incorrect attribution and loss of clinical detail. Consultations involving interpreters, several participants or complex narratives may require additional checking. Avoiding the tool may sometimes be appropriate when reliable documentation cannot be assured.
Medicolegal concern was generally modest, but it was not absent. Most users expressed at least some concern, and a small minority were highly concerned. Responsibility for the final clinical record remained with the clinician. Using an AI scribe therefore did not remove the need to verify documentation before it entered the patient record.
Consent practices also varied. Around two-thirds of users said they routinely asked patients before using an ambient scribe, while the remainder did not. Where consent was sought, refusals were uncommon. Notification and an opportunity to decline can therefore be incorporated into routine consultations, although patient views were not directly assessed.
Consistent consent procedures, systematic monitoring and evaluation across patient groups are needed. Reported effects on workload, note quality and error rates came from clinicians’ own assessments rather than independent observation. Prospective studies involving several sites and patient-reported outcomes would provide a fuller view of safety, acceptability and long-term use.
Ambient AI scribes offer clear practical benefits for some early users in primary care, particularly by reducing documentation time and mental workload. Their value nevertheless depends on close clinician review because errors remain frequent and correction is often required. Performance appears weaker in complex, multilingual and multiparty consultations, where failures may carry greater clinical consequences. Variation in consent practices adds another governance concern. Safe adoption requires reliable verification, transparent patient notification, monitoring across consultation types and independent assessment using objective safety, workflow and patient-reported measures.
Source: BMJ Health & Care Informatics
Image Credit: iStock
References:
Blease C, Kharko A, Sanchez CG et al. (2026) Ambient AI in primary care: an exploratory mixed methods survey of UK general practitioners. BMJ Health & Care Informatics;33:e101847.
