More than 4,800 Cleveland Clinic clinicians used an ambient AI scribe across 3.5 million patient encounters during the first year of an enterprise-wide deployment, according to a perspective article published in npj Health Systems.

The technology uses a large language model to convert a patient-clinician conversation into a draft encounter note. With patient consent, it records the conversation and transfers the generated note to the electronic health record, where the clinician is responsible for reviewing, editing, and signing it.

Amy Merlino, Chief Medical Information Officer at Cleveland Clinic, and colleagues described the health system’s deployment of the technology in partnership with Ambience Healthcare. Cleveland Clinic comprises 23 hospitals and 276 outpatient locations and employs approximately 80,000 people worldwide.

How the Rollout Worked

Following an internal evaluation of several AI scribe products, the health system began its enterprise ambulatory rollout on March 10, 2025. More than 4,000 clinicians, representing approximately 80% of Cleveland Clinic’s ambulatory workforce, were onboarded in less than 16 weeks.

Across the pilot and deployment phases, 6,130 clinicians were eligible for access. Of these, 5,748 (93.8%) completed training on the tool. Among those trained, 4,826 (84%) used the system at least once.

The percentage of clinicians who progressed from training to use declined across the successive deployment waves. All 104 clinicians trained during the pilot used the system. In the initial transition wave, 88.7% of trained clinicians used the AI scribe tool, 85.2% during the largest deployment wave, 82.6% in the second specialty wave, and 72.6% in the final wave, which included complex specialties where the AI models were still being optimized.

Among established users, defined as clinicians who had used the system at least 50 times, the AI scribe was used in 70% of eligible encounters. Use varied substantially by specialty, including among clinicians involved in cancer care.

Adoption in Cancer Care

By specialty, medical oncologists used the scribe in approximately 62% of eligible encounters, hematologists used the tool in about 65% of cases, and radiation oncologists used the scribe in about 23% of encounters. The analysis included 91 medical oncology clinicians, 81 hematology clinicians, and 24 radiation oncology clinicians. The article did not explain the differences among these specialties or report whether visit length, treatment complexity, documentation format, or other workflow factors impacted use.

The system may have particular relevance to oncology visits, which can involve extensive histories, pathology and imaging findings, molecular test results, treatment discussions, and coordination across specialties. However, the report did not assess whether the generated notes accurately captured cancer stage, biomarker findings, laboratory values, pathology results, or treatment plans. It also did not provide separate satisfaction, documentation-time, or safety results for oncology users.

Across all established users, Cleveland Clinic reported a Net Promoter Score of 60 and a customer satisfaction score of 96.6%. In addition, 60% of surveyed users agreed or strongly agreed that the scribe had increased their likelihood of remaining in clinical practice. The article did not provide a control group or predeployment comparison for these measures.

Operational Lessons

The authors attributed the scale of the rollout to four operational components: multidisciplinary governance, mandatory training, readily available support, and continuous feedback. The health system also established cross-functional stakeholder groups to support the vendor-system partnership and the project. 

Clinicians had to complete training before gaining access. Live virtual sessions were offered three times daily, supplemented by departmental sessions and self-directed modules. Department leaders received dashboards showing training progress and use within their specialties.

Support was available through a mobile chat service staffed by human personnel around the clock. More than 900 inquiries were handled during the rollout, with an average response time of 2 minutes. Common requests concerned physical examination templates, workflows involving multiple clinicians, corrections to draft content, and Cleveland Clinic policies.

Feedback from support channels, office hours, and a clinician advisory group was used to adjust specialty-specific note content, formatting, and settings. Later deployment waves focused on specialties with more complex workflows or less mature AI-generated outputs.

What the Study Did Not Measure

The article provides operational measures of adoption and use but does not establish clinical effectiveness. It did not evaluate patient outcomes, diagnostic accuracy, note completeness, error rates, safety events, or whether clinicians spent less time documenting. Patient experience and consent rates were also not reported.

For laboratory professionals, an additional unanswered question is how reliably ambient documentation captures and contextualizes laboratory, pathology, and molecular findings. Clinician review remains necessary because an inaccurate summary of a test result could affect communication among members of the care team even when the original laboratory report remains available elsewhere in the medical record.

Cleveland Clinic and the vendor are continuing to monitor ambulatory use and are developing approaches for emergency, inpatient, and nursing workflows. The report presents the rollout as an implementation framework rather than an independent assessment of the technology itself.

DISCLOSURES: The work received no external funding. Six authors were Ambience Healthcare employees with equity in the company. The Cleveland Clinic authors declared no competing financial or non-financial interests.

ASCO AI in Oncology is published by Conexiant under a license arrangement with the American Society of Clinical Oncology, Inc. (ASCO^®). The ideas and opinions expressed in ASCO AI in Oncology do not necessarily reflect those of Conexiant or ASCO. For more information, see Policies.