What is an AI scribe, or ambient documentation?
The AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
An AI scribe helps prepare clinical documentation from an encounter. An ambient system captures a conversation, processes the speech and generates a draft note, sometimes with other outputs such as patient instructions. The precise features vary by product and configuration.
This is more than transcription. Transcription attempts to record what was said. A clinical note selects and organizes information, distinguishing symptoms, history, assessment and plan. Errors can enter at either stage. A correct transcript can be summarized incorrectly, and a polished note can conceal a transcription mistake.
The attraction is straightforward: less time spent composing documentation and more attention available during the encounter. Evidence should nevertheless be described carefully. Studies differ in product, specialty, integration, follow-up period and outcome. Some report reductions in documentation burden or improved clinician experience; these findings do not establish the same benefit for every system or setting.
For example, a randomized trial published in 2025 compared two ambient tools with usual practice and found different effects on note-writing time. That variation is why headline claims that scribes save a fixed percentage of time rarely hold across every clinic. Self-reported well-being, measured EHR time and patient outcomes are also different endpoints. Improvement in one should not be used as evidence for all three.
Review remains substantive work. A draft may incorrectly record something as performed, omit an important qualification or attribute a family member’s history to the patient. Clinicians should be able to correct the note before signing and understand where the content came from. The organization also needs appropriate arrangements for informing patients, obtaining consent where required, and controlling recording, access and retention.
Evaluation should include language and accent variation, interruptions, multiple speakers and complex visits. Longer notes are not necessarily more useful notes. Patient instructions need to preserve the plan in language the intended reader can understand.
Finally, the service needs to decide how any time released will be used. A reduction in drafting work does not automatically reduce overall workload if other tasks expand to fill it. A good implementation examines the whole encounter and its follow-up, including the work of checking the draft, rather than treating fast note generation as the endpoint.
Sources: Lukac and colleagues — Ambient AI Scribes in Clinical Practice: A Randomized Trial (2025); A Pragmatic Randomized Controlled Trial of Ambient Artificial Intelligence to Improve Health Practitioner Well-Being (2025); Olson and colleagues — Use of Ambient AI Scribes to Reduce Administrative Burden and Professional Burnout (2025; quality-improvement study)
To learn more, take a look at the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy