An AI medical scribe is software that listens to the conversation between a doctor and a patient and writes a draft of the doctor's note. The doctor then reads the draft, corrects it and signs it. It is also called an ambient scribe, or ambient clinical documentation. "Ambient" means that it listens in the background, without the doctor dictating to it. Most of its buyers are doctors, clinics and hospitals in the United States.
Why is it needed? After every visit, a doctor must write a note: what the patient said, what the doctor found on examination, and what was decided. A doctor who sees 25 patients a day writes 25 notes. For years, doctors did this in one of two ways. Some dictated the note into a recorder, and a medical transcriptionist, a person who types up recordings, typed it. Many transcriptionists worked in India. Other doctors typed the notes into the record themselves, often late into the evening. The AI scribe replaced much of both in about two years, and with it much of the transcription work that Indian back offices did for American doctors.
Say a family doctor in Virginia sees a patient with a cough that has lasted three weeks. At the start, the doctor asks the patient's permission and taps Record on a phone. They talk for 12 minutes. The scribe turns the speech into text through speech recognition, and works out which voice is the doctor's and which is the patient's. A language model then turns the conversation into a note in the usual four sections: what the patient reported, what the doctor found, the doctor's assessment, and the plan. It leaves out the small talk about the weekend. One minute after the visit, the draft is in the patient's ambulatory EHR. The doctor reads it, changes one word and signs it.
The scribe now does more than the note. It drafts the order for a chest X-ray, and the referral letter to a lung specialist. It suggests the medical codes for the diagnosis and for the visit, which the billing system uses to send the claim. It can also answer the doctor's questions, such as the right dose of an antibiotic, from clinical decision support and from the patient's record.
The doctor still checks everything. A language model can mishear a word, or write down something that was never said. So the scribe drafts and the doctor approves, a rule called human in the loop. The doctor remains responsible for what is signed.
Other industries use the same idea. Sales and support teams use call transcription and AI summaries to write up their phone calls.