AI Scribes Cut Documentation Time in Primary Care

Artificial intelligence scribes substantially reduce the share of primary care consultations devoted to clinical documentation in a controlled simulation. A study published in JAMIA Open assessed nine primary care physicians in Ontario, Canada, as they completed four simulated encounters with standardised patients, both with and without an AI scribe. Documentation accounted for just over a tenth of encounter time when the tools were used, compared with more than a third during manual charting. Total encounter duration did not differ significantly, while reviewing and editing generated notes added time after the clinical conversation. The findings support further assessment before routine clinical implementation. 

Simulated Encounters Test Documentation Workflows 

The evaluation used a realistic primary care setting equipped with an electronic medical record sandbox, several web-based AI scribes and multiple cameras to record physician behaviour. The simulated consultation room included an examination bed, seating area and workstation, allowing participants to complete encounters under controlled conditions while using familiar clinical equipment. Screen recordings captured interaction with the electronic medical record and the AI tools, while synchronised camera views documented typing, writing, navigation and provider-patient interaction. 

Must Read: AI Scribes Cut Workload but Need Oversight 

Four standardised patient profiles represented common primary care presentations of differing complexity. Two encounters involved simpler cases and two involved more complex health needs. Each physician completed two encounters using manual documentation and two using an AI scribe. During manual encounters, physicians could enter notes during or after the consultation according to their usual approach. During AI-supported encounters, they reviewed the generated note, copied it into the record and made any edits they considered necessary before saving it. 

The tools generated notes in the Subjective, Objective, Assessment, Plan format. They operated outside the electronic medical record rather than through direct integration. Behavioural coding measured typing, handwriting, scrolling, copying, pasting, note generation and post-conversation processing. Each encounter was coded by two trained reviewers, with only observations meeting the predefined reliability threshold included. Time measures were converted into proportions of total encounter duration to account for differences in physician practice style, patient factors and communication. 

Documentation Falls While Review Time Rises 

Without an AI scribe, physicians spent an average of 36.3% of the simulated encounter on documentation. With an AI scribe, that share fell to 11.2%, representing a 69.1% reduction in documentation time. The reduction largely reflected lower typing time, while copying and pasting the generated note into the electronic medical record required little time. Charting measures also included any handwritten notes made during AI-supported encounters. 

The pattern changed after the clinical conversation ended. Post-conversation processing occupied 23.2% of encounter time with an AI scribe, compared with 16.2% without one. This measure covered the period from the end of the consultation to completion of the structured note in the record. The additional time was associated with reviewing and editing the generated content, either within the AI application or after transferring it into the electronic medical record. Some physicians also completed documentation after the conversation in line with their usual workflow. 

Despite the reduction in documentation, the average total encounter duration was not significantly different between the two conditions. Encounters were shorter on average when an AI scribe was used, but the difference did not reach statistical significance. Documentation practices also varied between physicians. Six of the nine participants had previous experience with AI scribes, while the group included clinicians from team-based, hospital, academic and solo practice settings. Most served substantial numbers of patients from historically underserved or equity-seeking populations. 

Further Evaluation Is Needed Before Adoption 

The findings are limited to documentation of structured clinical notes within simulated encounters. The evaluation did not measure time spent on referrals, billing, prescription orders or other administrative work. It also required physicians to complete all documentation in the consultation room before moving to the next case, which may not reflect workflows in which charting continues during administrative periods or after scheduled working hours. 

The standardised patient scenarios allowed direct comparison across participants but could not represent the full diversity of primary care presentations. Although the cases included simple and complex consultations, documentation savings may have been affected by case characteristics or differences between the AI scribes. Physicians also lacked access to electronic record shortcuts such as dot phrases or automated text, which can reduce manual documentation time. The extent to which this condition reflects routine practice is uncertain. 

The tools were separate web-based applications and were not integrated with the electronic medical record. Their use therefore required additional navigation and transfer of generated text. Integration may become an important factor in future assessments of usability and effectiveness. The evaluation also focused on time rather than the accuracy or quality of the generated notes. Further work is expected to examine clinician perspectives, note quality, administrative burden, burnout and performance in real clinical settings across more varied patient scenarios and physician workflows. 

AI scribes reduced documentation time during simulated primary care encounters, mainly by lowering the time physicians spent typing and charting. The tools did not significantly shorten total encounter duration and required additional post-conversation review and editing. The controlled setting showed how new documentation technologies can be assessed before routine use, but it did not capture the full range of administrative tasks or real-world clinical variation. Broader evaluations are needed to establish note accuracy, quality, usability and effects on workload across diverse consultations, electronic record configurations and working patterns. 

Source: JAMIA Open 

Image Credit: iStock