AI Scribes for Allied Health in 2026: What They Do, What They Don't and How to Choose

What an AI scribe is, and what it is not

An AI medical scribe listens to a clinical consultation, transcribes what was said, and structures it into a draft clinical note. That’s the full job description. You read the draft, correct what needs correcting, and sign off. The record is yours, and the responsibility for its accuracy is yours too.

Clinicians are looking at AI scribes because documentation takes too long. A 2016 study by Sinsky et al. (Annals of Internal Medicine) found US ambulatory physicians spent around 49.2% of their time on electronic records and desk work, against just 27% with patients. That’s a US physician study, not Australian allied health data, but the directional pressure will feel familiar if you’re still writing notes at 7pm.

An AI scribe is not a clinical decision-maker, a billing tool, or something that lodges anything on your behalf. It produces a draft. You produce the record.

What AI scribes can’t do, and where they fail

Errors happen, and the most common failure type isn’t a fabricated detail: it’s an omission. Something said in the room that the AI simply didn’t capture. You won’t always notice it’s missing until you read the note carefully.

Accents trip most tools up. So do multi-speaker environments: a family member who joins mid-session, a colleague who asks a question during the consultation. The AI often loses the thread. If you work with clients from diverse linguistic backgrounds, test the tool with those voices before subscribing.

Edit time erodes the time saved. A 2025 randomised controlled trial published in NEJM AI (reported via STAT News) found that time savings per note were modest, well below what vendors typically claim. More consistent across participants was a reduction in cognitive load and after-hours documentation. This is worth knowing, because “saves you two hours a day” is a common pitch with thin evidence behind it.

The clinician reviews and corrects every note before it enters the record. This isn’t optional; it’s your professional obligation under AHPRA guidance.

Using an AI scribe creates obligations that don’t apply to typing your own notes. AHPRA’s 2024 guidance on AI in healthcare is explicit: check accuracy, seek informed consent before recording or inputting client data to an AI, inform clients AI is being used, and the practitioner remains responsible for the record and the care provided.

Consent to record varies by state. NSW, SA, WA, Tasmania and the ACT require all-party consent before a conversation is recorded. Victoria, Queensland and the NT require only one-party consent (which can be you). AHPRA’s professional consent obligation sits on top of state recording laws regardless of where you practise.

Client audio and clinical notes are sensitive health information under the Privacy Act 1988. APP 11 requires reasonable security measures. The OAIC has noted that entering sensitive information into publicly available AI tools carries risk, particularly where server location and offshore disclosure aren’t transparent.

The TGA question

The TGA’s current position is that transcription alone is not a regulated medical device. A tool that independently generates a diagnosis, differential, or treatment recommendation may meet the definition and require ARTG inclusion. Ask any vendor you’re evaluating for their written TGA classification position. If they can’t provide one, that indicates a potential issue.

Run through these before you commit:

Accuracy and editability: Can you quickly find and fix errors? A tool that buries the source transcript makes review harder, not easier.
Australian data residency and APP compliance: Where does client audio go? Is it processed offshore, and under what conditions can that data be accessed?
Consent workflow: Does the tool support your consent process, or do you manage that separately with your own forms?
PMS integration: Nookal and Cliniko are the dominant practice management systems for Australian allied health. Native integration matters; copy-paste workflows add friction and create version-control risks.
Note formats: Does the tool produce physiotherapy and exercise physiology note formats, or does it default to GP SOAP? These aren’t interchangeable.
Scope beyond transcription: Does the tool do anything beyond producing a note? If so, know exactly what it infers versus what it transcribes directly.
Real edit load: Ask for honest average edit-time data, not headline time-saving figures.

AI clinical documentation: scribe vs clinical assistant

A scribe transcribes and structures a note; that’s a defined, bounded task. A clinical assistant does more: treatment plans, exercise prescriptions, referral letters, or funding documentation generated from the same session. These are different products with different risk profiles, and it matters which one you’re buying.

The further a tool moves from verbatim transcription, the more its output depends on inference. The more inference, the more carefully you need to review. That’s not a reason to avoid broader tools; it’s a reason to understand their scope before you rely on them.

Kitt sits at the clinical-assistant end of this spectrum. Beyond the session note, Kitt generates treatment plans, exercise prescriptions, and GP or referral letters from the same consultation. It also drafts documentation for NDIS, DVA, and workers compensation reports. The clinician reviews and approves every output before it’s used; Kitt doesn’t file, lodge, or submit anything on your behalf.

Kitt is Australian-built, hosted in Australia, and built to the Australian Privacy Principles. It integrates natively with Nookal and Cliniko. Notes generate in under 60 seconds. Like every AI tool, it can make errors, and you need to review output before it enters the record. Pricing is $69 per clinician per month, with a 30-day free trial and no lock-in contract.

For a direct comparison with another tool in the market: Kitt vs Heidi.

FAQs

Yes. AHPRA’s 2024 guidance requires informed consent before recording or inputting client data to an AI, and clients should be told AI is being used. State recording laws add a separate layer: all-party consent is required in NSW, SA, WA, Tasmania and the ACT.

Are AI scribes accurate enough to trust?

They produce useful first drafts, but omissions are the most common failure type. Read every note before it enters the record.

Is my client data safe if it goes offshore?

Client audio and notes are sensitive health information under the Privacy Act 1988. Offshore processing may constitute a disclosure under APP 8. Ask the vendor directly, and get the answer in writing.

Is an AI scribe a medical device?

Pure transcription tools are not regulated devices under the TGA’s current position. A tool that independently generates diagnoses or treatment recommendations may require ARTG inclusion. Ask your vendor for their written classification position.

If the AI makes an error in a note, who is liable?

You are. AHPRA is clear that the practitioner remains responsible for the accuracy of the clinical record and the care provided. The AI is not a registrant.

Key Takeaways

An AI medical scribe produces a draft clinical note; you review, correct, and own the record.
Time savings are real but smaller than vendor marketing suggests; reductions in cognitive load and after-hours work are more consistent than appointment-time savings.
AHPRA requires informed consent, accuracy checks, and disclosure to clients; state recording laws vary, so confirm the rules in your state.
Before committing to any tool, verify data residency, APP compliance, PMS integration, and TGA classification position.
Scribes and clinical assistants are distinct products: broader scope means more inference, which means more careful review.