Abridge and AI Clinical Documentation

Abridge is a company that builds ambient clinical documentation, a category of software sometimes called an AI scribe. During a medical visit, with the patient's permission, the software records the conversation between clinician and patient. Afterward, it turns that recording into a draft of the clinical note, the formal record of symptoms, findings, assessment and plan that clinicians must write for every encounter. The clinician reviews, edits and signs the note, which is then filed in the electronic health record (EHR). The aim is to let clinicians look at the patient instead of a keyboard and to reduce the hours spent finishing notes after the clinic closes. Abridge's CEO and cofounder, Shiv Rao, is a practicing cardiologist, according to Fortune.

The process works in stages. First, the clinician starts a recording on a phone or in the EHR once the patient has agreed. Second, automatic speech recognition, which Abridge says is tuned for medical conversations, cross-talk and background noise, converts the audio into a transcript. Third, summarization pipelines that combine large language models with Abridge's own datasets and methods turn the transcript into a structured note organized into standard sections. Fourth, Abridge says it ties each span of the summary to supporting evidence in the conversation, so a clinician can check any sentence against the transcript and audio. Finally, the clinician edits and signs, and the note, along with other outputs such as patient summaries and billing-relevant detail, is placed in the chart.

Abridge's growth tracks the rapid adoption of this category. In August 2023 it became the first company in Epic's new Pals program, the developer program that replaced Epic's App Market, which allowed its tool to work inside Epic workflows. On August 14, 2024, Kaiser Permanente announced it was making Abridge available to clinicians at its 40 hospitals and more than 600 medical offices, after testing and quality review. Fortune reported in July 2025 that Abridge raised a 300 million dollar Series E led by Andreessen Horowitz with Khosla Ventures participating, at a 5.3 billion dollar valuation, up from 2.75 billion at its Series D, and that it was used in more than 150 enterprise health systems. Abridge's own website now says it is used by more than 300 health systems and powers more than 100 million conversations a year.

The strongest published evidence on outcomes is a quality improvement study in JAMA Network Open, published October 2, 2025, led by Kristine D. Olson of Yale School of Medicine. It surveyed ambulatory physicians and advanced practice clinicians at six US health systems before and 30 days after they began using the same version of the Abridge scribe, from February to October 2024. Among 263 clinicians analyzed, burnout fell from 51.9% to 38.8%. Clinicians also reported lower note-related cognitive load, better focused attention on patients, and 0.90 fewer hours per week of after-hours documentation. The intervention was funded by the participating institutions, and Abridge provided pooled survey data but had no role in design or analysis.

Those results come with real limits that the authors spell out. There was no control group, so changes over time or enthusiasm for a new tool could explain some of the improvement. Participants volunteered and responses were not anonymous, which may favor early adopters and positive answers. Outcomes were self-reported rather than measured from EHR activity logs, one site was excluded from the burnout analysis, and several authors disclosed ties to Abridge, including a current senior director. Customer-reported figures on Abridge's website, such as an 86% reduction in documentation effort at one health system, are company marketing claims rather than independent studies.

Compared with alternatives, ambient AI sits between older approaches. Clinicians can type notes themselves, which is accurate but time-consuming. Dictation software transcribes what the clinician says word for word but still requires them to compose the note. Human medical scribes, in the room or remote, can produce good notes but are expensive and hard to staff. Ambient AI tries to provide a scribe's output at software cost. Abridge's main differentiators, by its own account, are tight EHR integration and its traceability from note text back to the source conversation.

The risks are those of any generative AI applied to medical records. A language model can omit something important, attribute a statement to the wrong speaker, or confabulate, meaning produce plausible text that was never said; Abridge itself has published work on confabulation elimination, which acknowledges the problem. Because a signed note becomes a legal and clinical record, the clinician remains responsible for its accuracy, and review takes real attention. Recording patient conversations raises privacy and consent issues, which is why Kaiser Permanente requires patient agreement and Abridge states that it encrypts data and complies with state and federal privacy laws. Accuracy can also vary with accents, languages, noisy rooms and specialties.

In practice, adopting Abridge is an enterprise decision rather than an individual purchase. Health systems contract with the company, connect it to their EHR, set consent workflows, train clinicians, and monitor quality, often starting with pilots, as Kaiser Permanente did with testing before its 2024 rollout. Abridge also sells revenue-cycle features that aim to make documentation billing-ready, which links the product to coding and reimbursement and calls for compliance oversight. For patients, the visible change is usually a consent request at the start of a visit and, in some cases, a plain-language visit summary afterward.