Getting Comfortable With AI Scribes
A plain-English guide to what an AI scribe actually does in the consulting room — consent, onshore data, model training and why clinician review of every note is not optional.
First published by In AI Today. Republished with permission.
A recent story in The Guardian described a Melbourne psychiatrist who told new patients they would need to consent to an AI scribe taking notes during their sessions, or find another practice. The reaction it drew said a lot. Much of the unease people feel about consenting to an AI scribe comes down to not really knowing what these tools are, how they work, and what happens to the conversation once the appointment ends.
So, it is worth walking through the questions people actually ask, in plain terms.
What actually is an AI scribe?
An AI scribe is software that listens to a consultation and turns it into a written clinical note. It records the conversation, produces a transcript, and drafts the note in the format a clinician would normally write up by hand. Most run quietly in the background on a phone or computer in the room. The best known in Australia are Heidi Health and Lyrebird, with others including PatientNotes, i-scribe and Telstra Health Smart Scribe. Between them, Heidi and Lyrebird account for roughly 95 per cent of the GP scribe market, and Heidi is the largest single player, with the company reporting that around half of Australian GPs now use it.
These are not the same as the note-takers built into Zoom or Microsoft Teams. A medical scribe is trained on clinical language, so it recognises the right terminology and structures the note the way a practitioner actually needs it, rather than producing a generic meeting summary. Uptake has been quick. GP use of AI scribes more than doubled in a recent four-month stretch.
If a practitioner uses an AI scribe, are they just outsourcing their skills?
This is the worry I hear most, and the answer is no. The scribe does not diagnose, decide or treat. It records the conversation and drafts the notes, and nothing more. The clinical thinking still belongs to the practitioner, and the typing is the only part being handed off.
Under the guidance issued by AHPRA and the RACGP, the responsibility sits squarely with the clinician. They are expected to obtain the patient’s consent before an AI scribe is used, and to review, correct and approve every note before it goes into the medical record. The scribe produces a draft which may be edited. The doctor or physiotherapist signs off on the record, exactly as they always have.
Where is the data held, and is it being sold?
For health information the clear expectations in Australia is that data stays onshore, and the major scribes have built for that. Both Heidi and Lyrebird state that patient data is stored on Australian servers. Both also say they do not sell patient data, and doing so would put them in breach of the Privacy Act in any case.
Is my patient data being used to train their model?
The same question comes up about model training, and the major Australian providers give the same answer. Heidi and Lyrebird both state that they do not use patient data to train their models. As with data storage, this is worth checking against the specific vendor’s current terms rather than taking it on trust, because the detail varies across the smaller players in a market that is still moving fast.
Why are practices moving to AI scribes?
The pull is mostly about time. In many of the practices we speak with, practitioners were routinely staying back two to three hours after their last patient to write up notes and chase admin, which also pushed out how quickly patients heard back. An AI scribe takes most of that load off, and the practice runs more smoothly without the extra overhead.
The part that surprises people is that it often improves the consultation itself. When the notes are being handled in the background, the practitioner can look the patient in the eye and listen attentively, rather than sitting side-on and typing into a computer. More attention on the person and less on the paperwork is why a number of clinicians who were sceptical have come around.
What are the risks, and how are they managed?
None of this means the tools are flawless, and it is right to be careful. AI scribes can make mistakes a human minute-taker would not. They can add detail that was never said, or leave out something that matters. Studies put the overall error rate at a low level, in the order of a few per cent, but the failure modes are different from human ones, which is exactly why clinician review of every note is not optional.
There is also a longer-term risk worth naming. If a fabricated or wrong detail slips through unreviewed, it can be carried forward from note to note and quietly harden into accepted history, so the review step protects not just today’s record but every consultation that builds on it.
The other risks are about consent and security. Everyone in the room has an interest in being recorded, so consent should extend to family members or carers who are present, and be refreshed when someone new joins. And because the recording captures sensitive health information, it needs to be stored and transmitted securely, which is another reason to choose an Australian-hosted provider and to keep the consent conversation front and centre at every visit rather than treating it as a one-off form.
Handled properly, an AI scribe hands the clinician back time and can improve the patient experience. As use of AI scribes becomes ubiquitous, it is worthwhile understanding what they are so that you do not need to fear them.