Your record-keeping
obligations

An AI scribe can draft the note, but the record still belongs to the clinician who reviews and signs it.

An AI scribe drafts the note. The registered practitioner who reviews and signs it is still the one responsible for what it says, and that doesn't change no matter how accurate the transcription is or how confidently the draft reads.

Close-up of a person's hands in a blue shirt signing paperwork with a pen, documents spread across a dark desk

The obligation doesn't move with the tool

An AI scribe changes how a note gets written. It doesn't change who the record belongs to under professional regulation. The practitioner named on the consult record, the one whose sign off finalises it, carries the same accountability they always have: for accuracy, for completeness, for whether the note reflects what was actually discussed and decided. A transcript is not a record. A reviewed, signed note is.

This distinction matters more as scribing tools get better at sounding right. A fluent, well structured draft is easy to trust on sight, and that is precisely the habit worth resisting. AHPRA's registration standards and the shared Code of conduct that the National Boards administer don't carve out an exception for machine assisted documentation. The obligation to keep an accurate, contemporaneous clinical record sits with the registered practitioner, whatever tool produced the first draft.

What the record actually has to do

Clinical records serve several audiences at once: the next clinician who reads them, Medicare compliance reviewers checking that a consult supports the item billed, a medical board investigating a complaint, and, occasionally, a court. The Medicare Benefits Schedule requires a contemporaneous, clinically relevant record for the service billed, not a summary reconstructed from memory weeks later. The RACGP's Standards for general practices sets out expectations for what an accredited practice's clinical records should capture: presenting problem, relevant history, examination findings, working diagnosis, management plan, and follow up.

An AI drafted note has to meet the same bar as a typed one. It can meet it well, because a scribe captures detail a rushed clinician might otherwise compress into a one line summary, but meeting the bar is still the clinician's job to check, not the tool's job to guarantee.

Review before finalising is where responsibility attaches

The moment a clinician reads a draft note and confirms it before it becomes part of the patient's record is not a formality bolted on to satisfy a vendor's compliance page. It's the point at which the content becomes the practitioner's own professional judgement rather than a machine's best guess at what was said.

The profession already has a mental model for this: a consultant reviewing a registrar's letter before it goes out under their name, a GP checking a locum's note before continuing a patient's care, a specialist correcting a typist's transcription of a dictated letter. An AI scribe sits in the same place in that chain. The clinician reviews, edits where needed, and only then does the note attach to the record. Skipping that step, treating the draft as final because it reads well, is where the risk actually lives, not in the fact that a machine helped write the first pass.

Where an AI drafted note goes wrong, and what review is actually catching

The failure modes worth checking for are specific, not vague concerns about AI in general. A clinician who knows what to look for can review a drafted note in under a minute.

None of these are exotic edge cases. They are the reason the review step exists, and why the scribe writes, the doctor decides is the operating principle rather than a marketing line.

  • Medication names and doses that sound similar in speech: a mis-heard dose is the single highest value thing to check before signing.
  • Negative findings the clinician noted mentally but didn't say aloud, so the scribe has no way to record them. A clear chest only appears in the note if someone said it.
  • Attribution errors in a consult with more than one speaker, where a comment from a parent, interpreter, or support person gets folded into the patient's own words.
  • A plan the scribe infers from context rather than one the clinician actually stated, particularly around follow up timing or referral urgency.

Questions worth settling before adopting any AI scribe

The record-keeping questions above apply to any tool a practice is considering, not just one vendor. A practice manager evaluating options should be able to get a straight answer to each of these before rollout, not after a complaint forces the question.

  • Where is audio and transcript data stored, and does that location matter for a practice bound by Australian privacy law.
  • How long is raw audio retained after a note is finalised, and can that retention period be shortened.
  • Who can see the audit trail showing what the scribe drafted versus what the clinician changed, and can that trail be produced for AHPRA, a medical defence organisation, or a court if a complaint arises.
  • Is patient consult data used to train the vendor's models, and if so, is that disclosed plainly rather than buried in a terms document.
  • Does the tool integrate with the practice's existing clinical software, such as Best Practice, MedicalDirector, Genie, or Cliniko, so the finalised note lands in the same record every other entry sits in, rather than a separate silo.

Where Aurii fits

This section is about our product. Everything above is not.

Aurii is built around the review step described above, not around removing it. It listens to a consult with the patient's consent, drafts the progress note, referrer or GP letter, and discharge summary, and then stops. Nothing becomes part of the record until a named clinician reads the draft and signs it off. That sequencing is deliberate: the scribe writes, the doctor decides.

Underneath that, the governance questions above have direct answers. Consult data is captured, transcribed, and stored in Australia, with Sydney as the primary location and Melbourne as backup. Each record is encrypted, and every access and edit sits in a seven-year tamper-evident audit trail, the kind of trail a practice would want on hand if AHPRA or a medical defence organisation ever asked how a note came to read the way it does.

Aurii is a documentation aid. It is not a medical device, it is not on the Australian Register of Therapeutic Goods, and it does not diagnose or recommend treatment. It drafts paperwork faster so the clinician's judgement, not the software's, is what ends up in the record. Practices considering it, or any AI scribe, should read that governance layer as closely as the transcription quality, since for AHPRA purposes it is the layer that matters more. See how Aurii's compliance and privacy settings are configured and the detail behind the storage, encryption, and audit trail.

Common questions

The registered practitioner who reviews and signs the note, not the software. That responsibility already sits with the clinician for anything typed, dictated, or transcribed by someone else, and an AI scribe doesn't shift it.

No. The Medicare Benefits Schedule still requires a contemporaneous, clinically relevant record supporting the item billed. A scribe can help capture that detail more completely, but the clinician still has to confirm the finalised note actually supports the service claimed.

A tool that transcribes a consult and drafts documentation, without diagnosing a condition or recommending a treatment, generally sits outside the Therapeutic Goods Administration's software based medical device framework. Aurii is built and positioned this way, as a documentation aid rather than a device on the Australian Register of Therapeutic Goods. Practices should confirm how any specific vendor is positioned before assuming the same applies.

Enough to reconstruct what the scribe drafted, what the clinician changed, and when the note was finalised and signed. If AHPRA, a medical defence organisation, or a court ever asks how a note came to read the way it does, that trail is what answers the question, which is why a seven-year tamper-evident record of access and edits matters more than transcription accuracy alone.

This article is general information about record-keeping practice and does not constitute legal, clinical or regulatory advice for any particular practice. More guides sit on the resources hub. If your practice needs a question answered before it adopts AI documentation, tell us and we will write it: hello@aurii.com.au.

Read first.
Then see it on your own round.

You do not have to take any of this on faith. Request access, bring a real consult, and watch the note, letters and discharge come out the other end, yours to correct and sign.

hello@aurii.com.au

Stay in the loop.

Leave your email and we'll be in touch. No spam, unsubscribe any time.