Background: Recent advancements in large language model (LLM) processing have ushered in an era of artificial intelligence (AI) based scribe services to assist clinicians with documentation. Various vendors now offer ambient tools that generate documentation by listening to patient encounters. It remains unclear what clinicians expect from these services, how they prioritize features, and what concerns they have as organizations move toward selecting a vendor. We surveyed clinicians at an academic medical center to determine their perspectives on ambient scribe features and to identify key considerations for leadership when choosing and implementing an institutional AI scribe vendor. The survey was conducted during a period when the institution was piloting multiple AI scribe vendors.
Methods: We conducted a mixed-methods cross-sectional survey among a convenience sample of clinicians at a single academic center including clinicians who had previously expressed interest in scribe issues, urology faculty, and members of the physician scribe advisory group. The survey was distributed via email and professional listservs to an estimated 100 clinicians, with the option for recipients to forward to colleagues. The 4-question Qualtrics survey was distributed between May 20, 2025 through June 16, 2025 and assessed current scribe use, most important scribe features, and perspectives on institutional vendor selection. Quantitative items were summarized descriptively. Open-ended responses underwent inductive thematic analysis by two independent reviewers who reconciled themes by consensus.
Results: A total of 69 clinicians responded to the survey. Ambiance was the most commonly used AI scribe, followed by Augmedix (Table 1). Accurate documentation was the most frequently selected “most helpful” feature (47/69, 68%) followed by pre-charting elements of the history of present illness (HPI) (33/69, 48%) and importation of clinical data into notes (e.g. labs, microbiology, radiology) (26/69, 38%). Qualitative analysis showed that clinicians valued functionality – including accuracy, pre-charting, test interpretation and template customization – as the key considerations for an institution-wide vendor. Additional themes included the importance of clinician input and feedback to drive iterative improvement, a wide range of positive and negative experiences, limitations of current scribe features, instability associated with shifts from human scribes to multiple AI vendors, and potential consolidation to a single system (Table 2).
Conclusions: The AI scribe features most valued by clinicians were accurate documentation, pre-charting elements of the HPI, and importation of clinical data. Clinicians emphasized functionality, including accuracy, customizability, and support for core documentation tasks, when assessing AI scribe options. Qualitative findings also highlighted variability in user experience, limitations of current systems, and the instability created by frequent transitions between scribe approaches. Although the sample was modest and drawn from a convenience cohort, these findings underscore the importance of incorporating clinician input as institutions move toward selecting vendors in a rapidly evolving AI landscape.
.png)
.png)