The clinician
still signs

An ambient scribe listens to a consult, with the patient's consent, and turns the conversation into a structured draft: a progress note, a referral or GP letter, a discharge summary. That draft is only useful if it is accurate, and accuracy here means two things at once. Everything clinically relevant that was said needs to appear, and nothing that was not said should appear either. Most people worry about the first and underweight the second.

A consult is rarely a clean audio sample. Overlapping speech, a patient trailing off mid sentence, a regional accent, background noise in a shared clinic space, and dense clinical shorthand all make transcription harder before drafting even starts. None of that is a reason to distrust the tool outright. It is the reason review exists as a designed step, not an afterthought.

Omission and fabrication are not the same problem and they do not carry the same risk. An omission, a detail the patient mentioned in passing that did not make it into the draft, is usually caught because the clinician remembers saying it and notices it missing. A fabrication is more dangerous precisely because it is not missing: it reads as a normal, plausible clinical sentence, sitting in the expected place in the note, with nothing about its phrasing to flag it as wrong.

Language models are built to produce fluent, plausible text. Where the underlying audio is ambiguous, the risk is a system that fills the gap with something that fits the pattern of a clinical note rather than something drawn directly from what was actually said. A dose that sounds standard, a finding that sounds typical, a plan that sounds sensible: all of these can read as polished and still be invented. This is why a scribe worth trusting is built to draft from the transcript itself rather than compose freely around it, and why how Aurii works underneath the surface matters as much as what it produces.

Under the Medicare Benefits Schedule, a clinically relevant record needs to be contemporaneous and accurate, and that obligation sits with the treating clinician, not with the software used to draft it. AHPRA's expectations of record keeping work the same way: the practitioner who provided the service is the one accountable for what the record says happened. A drafted note, however well drafted, has not met that bar until a named clinician has read it, corrected anything wrong, and signed it as their own.

This is also why an ambient scribe is properly understood as a documentation aid rather than a medical device. It does not diagnose and it does not recommend treatment, and it is not registered on the Australian Register of Therapeutic Goods, because it is not making clinical decisions. The clinician is, and review is where that decision gets exercised over the draft rather than assumed from it.

A useful review habit is not a full re-read of every sentence. It is a short, targeted check of the parts of a note most likely to carry an error and most likely to matter if wrong.

• Medication names, doses and frequencies, checked against what was actually prescribed or continued
• Allergies and known adverse reactions, confirmed present and unchanged from what the patient reported
• Any measurement or figure, observations, results, dates, checked against the source rather than assumed correct because it looks tidy
• The line between what was examined and what was planned, since a drafted note can occasionally blur a clinical finding into the management plan
• Follow-up and safety-netting advice, confirmed as what was actually said to the patient
• Anything that reads as more specific or more certain than the consult actually was

The point of an ambient scribe is to give clinicians back the minutes usually spent typing during or after a consult. That benefit disappears if review turns into a second draft. The habit that works is reviewing immediately, while the consult is still fresh, against the same short checklist each time, rather than a slow line by line read days later when nothing about the encounter is easy to recall.

The first few weeks with any scribe are effectively a calibration period. Every clinician's speech pattern, accent, and specialty vocabulary is different, and a scribe used consistently will show where it tends to need correction, whether that is a particular drug name, a shorthand phrase, or a specialty term used often. Once that pattern is known, review gets quicker, because the clinician knows exactly where to look rather than reading blind.

For a practice manager, review discipline should not depend on individual habit alone. A written policy that states who reviews a draft, that no note is finalised without a named clinician's sign-off, and that draft and signed versions stay distinct in the record protects the practice as much as the patient. A clinician joining the practice should be told plainly what signing a note means: it is not administrative approval, it is the clinician taking the same accountability for that note as if they had typed every word of it themselves.

Whether the practice runs Best Practice, MedicalDirector, Genie, Cliniko, or Halaxy, the review step belongs before the note is written back into that system, not after, so the record that lands in the patient's file has already been checked rather than merely generated. The same governance applies to anything later uploaded to My Health Record, operated by the Australian Digital Health Agency: it receives the clinician-approved record, not the draft.