AI for Australian allied health practices: what's allowed

How can allied health practices use AI in Australia?

Use AI for the administration around care, not for the care itself. Four workflows suit most physiotherapy, psychology, occupational therapy, podiatry and speech pathology practices: the front desk and bookings, intake form summaries, referral and report letter drafts, and session notes drafted by an AI scribe with the client’s consent. In each one the AI prepares something and a person checks it before it reaches a client or the record.

Workflow	What the AI does	What stays with a person	Rule to watch
Front desk and bookings	Answers admin questions, books into the practice system, sends reminders	Anything clinical, complaints, unusual requests	Collect only what a booking needs
Intake form summaries	Turns a long intake form into a one-page pre-read	Reading every risk answer in full, planning the first session	Health information is sensitive information
Referral and report letters	Drafts GP letters and progress reports from your notes	Clinical opinions, recommendations, the signature	The practitioner is accountable for the content
Session notes (AI scribe)	Writes up the consult in your note format	Consent, correcting every line, saving to the record	AHPRA consent guidance, TGA rules on suggestions

Start with the front desk: no clinical content passes through the AI. Scribes come last, once your consent wording and vendor checks are settled.

Which rules apply to AI in allied health?

Two layers apply. AHPRA’s guidance on AI binds practitioners in the registered professions, including physiotherapy, psychology, occupational therapy, podiatry, chiropractic, osteopathy and optometry. Speech pathology, dietetics and exercise physiology are self-regulated through their professional associations, so AHPRA’s guidance does not formally bind them. The Privacy Act, however, covers every health service provider, whatever its size.

AHPRA’s guidance. Meeting your professional obligations when using Artificial Intelligence in healthcare (22/08/2024) says registered practitioners:

• remain responsible for their practice and must apply their own judgement to anything an AI tool produces;
• must check the accuracy of records an AI scribe produces;
• should generally get informed consent before using an AI scribe, and note that consent in the record;
• must protect patient confidentiality and privacy;
• should understand whether patient data will be used to train the vendor’s AI model.

Self-regulated professions. Speech Pathology Australia, Dietitians Australia and Exercise & Sports Science Australia set their members’ codes. Check yours for AI guidance, and treat AHPRA’s five points as a sensible standard anyway.

The Privacy Act. Most businesses with annual turnover of $3 million or less are exempt, but health service providers are covered regardless of turnover, and health information is sensitive information. Victoria, NSW and the ACT also have their own health records laws. Since 10/06/2025, people can also sue for serious invasions of privacy under a statutory tort. Our overview of Australian AI compliance covers the other regulators.

Are AI scribes allowed for physios and psychologists?

Yes. No Australian law bans AI scribes, and AHPRA’s guidance treats them as a tool registered practitioners may use if they stay responsible for the record. In practice that means informed consent before the session, a scribe whose data handling you have checked, and reading and correcting every note before it is saved. Psychologists need extra care, because session content is often far more sensitive than a physiotherapy consult.

Australian scribe tools include Heidi Health and Lyrebird Health, and some practice systems now build a scribe in (see the Nookal row in the table further down). Whichever you consider, get written answers to these questions before a client session goes through it:

1. Where are the audio and the text stored, and does either leave Australia?
2. How long is the audio kept, and can one client’s recording be deleted on request?
3. Is our data used to train the vendor’s models, and is that off by default or a setting we must change?
4. Who at the vendor can access session content, and is that access logged?
5. Does the tool ever suggest a diagnosis or treatment, and if so, what is its ARTG entry?

Habits differ by profession. Physiotherapists should check objective measures and the home exercise program line by line. Psychologists should treat consent as ongoing: a client who agreed in week one can decline when a session turns to trauma or risk, so pause the scribe and write that note yourself. For children in paediatric occupational therapy or speech pathology, consent comes from the parent or guardian.

Do I need patient consent to use an AI scribe?

Generally yes. AHPRA’s guidance says practitioners should get informed consent before using an AI scribe and record that consent in the patient’s notes. Informed means the client understands that software is listening and drafting the note, what happens to the audio and text, whether anything is stored overseas, and that they can say no, or ask you to pause it, without it affecting their care.

A short script you can adapt:

I use an AI note-taking tool that listens to our session and drafts
my notes. The recording is [deleted once the note is written / kept
for X days], it's stored [in Australia / overseas], and it isn't used
to train the AI. I check and correct every note myself. You can say
no, or ask me to pause it at any time, and it won't change your care.
Are you happy for me to use it today?

Then:

1. Record the answer in the client’s notes each time, for example “Verbal consent to AI scribe, 14/10/2026”.
2. Add a line about the scribe to your intake form and privacy policy, so nobody hears about it for the first time in the room.
3. Keep a non-scribe way to write the note, so declining costs the client nothing.

Is an AI scribe a medical device?

Usually not. The TGA said in August 2025 that a digital scribe which only transcribes and writes up what was said is not a medical device. If the tool suggests a diagnosis or a treatment that the clinician did not state, it is functioning as a medical device and must be included in the Australian Register of Therapeutic Goods (ARTG) before it can be supplied.

So if your scribe offers to “suggest” a diagnosis, treatment or exercises you never mentioned, switch that feature off or ask the vendor for its ARTG entry. Either way, under AHPRA’s guidance the note and the clinical decision are still yours.

Can AI summarise intake forms safely?

Yes, if the summary is a pre-read and not a substitute for reading the form. The AI turns a long intake into a one-page brief: reason for referral, goals, relevant history, listed medications, funding type and anything the client flagged. Any answer about self-harm, suicide, violence or safety should be copied word for word to the top of the brief and read by a clinician the same day, never paraphrased.

Connect it to your intake form so the text never sits in a staff member’s personal AI account. The same set-up drafts GP letters and progress reports from your notes: you supply the clinical content, then edit and sign.

Can AI book appointments into Cliniko or Nookal?

Yes. An AI front desk connected to Cliniko or Nookal can read live availability, book the right appointment type with the right practitioner, and send confirmations and reminders without anyone at reception. It should collect only what a booking needs: name, contact details, appointment type, preferred practitioner and time. It should not ask about symptoms, diagnoses or history; the intake form and the clinician handle those.

What your practice system already includes matters, so check before you buy anything (as at October 2026):

Practice system	Price	Built-in AI
Cliniko	US$45 a month for one practitioner, billed in US dollars	No built-in AI scribe listed on its features page
Nookal	Essentials A$49, Professional A$59 per practitioner a month, ex GST	”AI + Voice” add-on (AI scribe, voice-to-text notes, AI letters) at $25 per active licence a month
Halaxy, Power Diary	See each vendor’s current pricing	Check what your plan includes

For practices on Halaxy or Power Diary, we check during the audit whether a direct booking connection is possible. Where it is not, the front desk takes the request and preferred times, and your team confirms it in the morning. Our guide to AI receptionists covers after-hours setup in more detail, and the AI for physiotherapy clinics page goes deeper on one profession.

Which AI tools are safe for client information?

Only tools covered by a written agreement on where data is stored, how long it is kept and whether it trains the model. The OAIC recommends, as a matter of best practice, that organisations do not enter personal information, particularly sensitive information, into publicly available generative AI tools. A personal ChatGPT or Claude account is the wrong place for anything that identifies a client.

• Consumer plans. ChatGPT Free, Plus and Pro may use chats for training unless you turn it off. Since 28/08/2025, Claude Free, Pro and Max users choose whether chats train Claude.
• Business plans and APIs. ChatGPT Business and Enterprise, Claude Team and Enterprise, and both companies’ APIs are not used for training by default.
• Where data is processed. Anthropic’s own API processes data outside Australia (globally by default, or US-only on request). Claude on Amazon Bedrock with the Australian cross-region profile keeps data in the Sydney and Melbourne regions for supported models. If data goes overseas, APP 8 makes you accountable for how it is handled.

Our AI privacy guide for Australian businesses compares the plans in more detail.

Where are the red lines for AI in an allied health practice?

Three things stay with clinicians, however good the tool: clinical advice, anything involving risk or crisis, and any clinical message to a client. AI can prepare material for all three, but it must never answer a clinical question, assess risk or send clinical content on its own. Set these rules before go-live.

• No clinical advice from the AI. A website or SMS assistant answers fees, hours, parking, what to bring and how to book. “Is this exercise safe with my knee?” goes to a clinician.
• Crisis and risk content always goes to a clinician. An assistant that sees distress gives 000 and Lifeline (13 11 14) immediately and alerts the practice. It never tries to counsel.
• Nothing clinical is sent without review. Booking confirmations and reminders can go automatically. Letters, reports, home program changes and anything that discusses a client’s condition are read and approved by the clinician first.

How On Autopilot can help

On Autopilot’s AI Front Desk gives allied health practices an assistant on website chat, SMS and email that answers admin questions, books into Cliniko or Nookal, sends reminders and passes anything clinical or urgent to your team. It never asks for payment details, and never asks for more health information than your own intake form already does. It costs $1,500 to set up plus $199 a month (ex GST) and takes 7–10 working days, including a shadow week. Managed AI, from $1,500 a month, adds intake summaries and letter drafting on a plan that does not train on client data. We do not sell scribes, but we will help you vet one. Start with a free 30-minute AI audit with Jenn Yang.