AI in Healthcare Administration: A Practice Guide

I ran my own multidisciplinary clinic in New Zealand. The tool that saved us the most time wasn't an AI product at all. It was swapping practice management systems, to one called Splose. It reconciled our ACC billing far faster, and it had voice-to-text note-taking built in. (I've worked with Splose in the past. Weigh what I say about them with that in mind.)

A practice management system is the software your clinic runs on: bookings, patient records, invoicing, all of it.

That's roughly where AI sits for a clinic in Australia or New Zealand right now. The work it does is admin. A form that fills itself in. An inbox that sorts itself. A consult note that drafts itself for someone to check.

One of those tools, the AI scribe, is getting picked up faster than the rules around it are settling. Get it straight before you buy anything else.

Two things get called "AI in healthcare"

One is clinical. A model reads a scan, flags an early sign of disease, helps with a diagnosis. That work is real. It sits with hospitals, researchers and device regulators like the Therapeutic Goods Administration (TGA). This guide isn't about that.

The other is operational. Admin, notes, patient messages, marketing and reporting. That's the one you'll actually be sold, and that's the one this guide covers.

AHPRA is the national body that registers health practitioners in Australia and sets the rules they work to. It gives five principles for using AI: accountability, understanding, transparency, informed consent, and the ethical and legal issues involved (AHPRA). Those apply to an admin guide like this one too, because a scribe sits close to the clinical record. The federal Department of Health, Disability and Ageing has its own AI in health care page as well. It points to an AI Clinical Use Guide built with the Australian Commission on Safety and Quality in Health Care.

Patients already draw this line themselves. RACGP-reported research found 53% of people are fine with AI handling appointment booking (RACGP, Health of the Nation 2025). Only 32% are comfortable with AI-assisted diagnosis, and 76% still expect a clinician to check anything AI suggests (RACGP, Health of the Nation 2025).

How fast AI is actually moving in AU and NZ practices

An AI scribe is software that listens to a consult and writes up the note, which a clinician then reads, fixes and signs. It's the clearest evidence of how fast this is moving.

RACGP runs its own newsGP poll. Scribe use among GPs rose from 22% in August 2024 to 40% by November 2025, close to double in fifteen months (RACGP's newsGP). The Guardian later reported the same figures (The Guardian).

newsGP also carried numbers from Avant, a medical defence organisation. Scribe use among its members went from 11% in August 2024 to 19% in February 2025 (RACGP's newsGP). Different group, smaller number, same direction. The same newsGP article reported a JAMA-published study of 186 clinicians. Their self-reported admin burden dropped from 52% to 39% after a month with a scribe (RACGP's newsGP).

Now set that against RACGP's Health of the Nation 2025 survey of just over 3,000 GPs (RACGP, Health of the Nation 2025). At least 80% said they were not at all, or not very, familiar with specific AI tools. AI-powered clinical decision support was the least familiar of the lot, at 9% (RACGP, Health of the Nation 2025). Use of one tool is climbing fast. Understanding of the category isn't.

Scribes are built to be easy. Someone else sets it up, it takes a few minutes, and nobody has to learn how AI works or where else it might help. That ease is the whole design. It's also why the understanding gap exists.

The consent gap regulators are already worried about

The Guardian got hold of federal Health Department briefing papers under freedom of information, written for Senate estimates. They said AI scribes "have little oversight". They also said some vendors had advertised "a 30% revenue increase for health professionals with no additional hours or patient consultations" (The Guardian, 5 July 2026).

The same reporting said some cloud platforms send data outside Australia without the supplier knowing. It quoted the Australian Privacy Commissioner, Carly Kind, saying her office is "tracking closely" how scribes are rolling out, and has found "deficiencies in GPs' implementation of AI scribe consent protocols" (The Guardian).

RACGP's own fact sheet on scribes covers patient consent too. It notes that some medical defence organisations tell you to get it in writing (RACGP's AI scribes fact sheet). Our guide to choosing an AI scribe walks through the whole thing. AHPRA's informed consent principle is the same point from the regulator's side (AHPRA).

AI in healthcare administration

Most of what a practice manager calls AI in healthcare administration is the same repetitive work, done a bit differently. For example:

• A new patient fills in an intake form online. AI checks it for gaps and drops the details into the practice management system, so nobody retypes a Medicare or NHI number.
• The inbox that used to eat an hour every morning sorts itself. Booking requests go to the calendar, billing questions to the front desk, and the rest waits for a person.
• A missed call after hours gets a message back offering the next free slot, instead of that caller ringing the clinic down the road.
• Referral letters and forms get scanned and filed against the right patient record, instead of sitting in a stack until someone has time.

Before you buy any of it, know what the pitch leaves out. I've sat on both sides of that conversation. First buying for a clinic, later selling into them. I've watched vendors get excited and oversell before they understand the privacy and data concerns healthcare carries. They also underestimate how hard it is to connect systems, especially older ones. Sometimes the API access to move data between them doesn't even exist. An API is the doorway one piece of software opens so another one can read its data.

The practice manager still runs the desk. AI clears the paperwork underneath it.

AI and clinical documentation: where the line sits

This is the one place in this guide where AI touches the clinical record.

The TGA's rule is clean. A digital scribe that only transcribes a consult isn't a medical device. One that goes further, and gives you a diagnosis or a treatment recommendation, is. That one has to be listed on the Australian Register of Therapeutic Goods (TGA, Digital scribes).

What a scribe hands you is a draft. You read it, correct it and sign it, the same as you would a note from a registrar. It saves the typing. The judgement stays with whoever signs.

RACGP's fact sheet ends with a list of questions to ask before you pick a scribe. Where is the data stored, for one. Does the vendor use it for anything else (RACGP's AI scribes fact sheet). Work through it before the demo, and get the answers in writing.

AI in patient communication

Patient care doesn't only mean clinical care. Most of what a patient actually experiences with your clinic is admin. Booking, reminders, follow-up, waiting.

In practice, that looks like this:

• Appointment reminders and recall messages going out on a schedule, with a person approving anything before it sends.
• A chat widget on the website answering "do you bulk bill", "are you open Saturday" and "where do I park", so the front desk isn't fielding the same three questions all day.
• Follow-up messages or care instructions after an appointment, once the practitioner has okayed the wording.
• A cancelled slot offered straight to the next patient on the waitlist, instead of sitting empty until someone notices.

The aim is fewer patients falling through the gaps between appointments.

Generative AI in healthcare marketing

Generative AI is the kind that writes or makes something new, rather than sorting what's already there. In marketing it's mostly a drafting tool. It does the first pass, and a person finishes the job.

Consider a few of the jobs it's genuinely good at:

• A first draft of a social post about a new service, written from a few dot points a practitioner gives it, then edited by someone who knows how the clinic sounds.
• A monthly newsletter drafted from the clinic's own updates, instead of written from a blank page every time.
• Website copy drafted for review, so a redesign doesn't stall on an empty screen.

A person still checks tone, accuracy and anything that reads as a health claim before it goes out.

Privacy and data: what AU and NZ law actually require

In Australia, the OAIC has been blunt about it: "Privacy obligations will apply to any personal information input into an AI system, as well as the output data generated by AI (where it contains personal information)" (OAIC). What goes in is covered. What comes back out is covered too. The Australian Privacy Principles stop you using personal information for anything beyond what you collected it for. OAIC says run a Privacy Impact Assessment before you bring a new AI tool in.

New Zealand's Privacy Commissioner got to the same place on its own. The Information Privacy Principles under the Privacy Act apply to AI tool use, and a Privacy Impact Assessment is recommended there too (NZ Privacy Commissioner). The Medical Council of New Zealand signed off its own AI guidance for doctors in March 2026, the NZ counterpart to AHPRA's (MCNZ).

Health New Zealand goes further again with its own staff and contractors: "the group recommends caution when using these tools within Health NZ. We need more data to properly evaluate the risks and benefits in the New Zealand health context" (Health NZ). Staff are told not to put "any personal, confidential or sensitive patient or organisational information into unapproved LLMs or generative AI tools". They're also told not to use them "for any clinical decisions, or for personalised advice to patients." That's written for Health NZ's own workforce, not for a private clinic. The caution is still worth borrowing for any tool that touches patient data.

Where to start, and how Opal can help

Start before AI. Start with what's costing you patients or hours right now.

Say your biggest goal this quarter is new patients. The first thing worth checking isn't an AI tool. It's the basics. A website that's actually set up to be found. A Google Business Profile that's filled in and correct. Honest Google reviews from patients. Asking for a review is fine. Offering anything in return isn't, because Google's own policy bans incentivised reviews (Google). If the basics are off, nobody is seeing your clinic, and no amount of AI fixes that.

AI helps once the fundamentals are in place. Drafting website content, for instance, or improving how you show up in Google and in AI search. Often that needs nothing more than the chat window in a general assistant such as ChatGPT, Gemini or Claude. That part is easy to start.

The two mistakes I see most often run the other way. The first is hype beating the outcome. A clinic connects its accounting software for better reporting when the real goal this quarter is new patients through the door. Useful one day. Not the thing moving the number that matters now. The second is underestimating the complexity, so the project stalls halfway through once the hard part shows up, with nobody involved who has done it before.

We'd start with an audit. Which gap in the patient experience, or which report you can't currently see, is costing you the most hours or the most patients. That's the starting point for AI strategy and adoption, reporting included, so you can see whether a change worked. We don't sell software, and we take no commission on what we recommend.

FAQ

What are examples of AI in healthcare?

In a clinic the clearest examples are AI scribes drafting consult notes and chat widgets answering after-hours questions. Then inboxes that triage themselves, and tools that draft a first pass of a newsletter or some website copy. The broader examples, such as AI reading a scan, belong to hospitals and researchers rather than a typical clinic.

Is AI used in Australian clinics?

Yes, and for at least one tool it's growing fast. RACGP's newsGP poll found scribe use among GPs rose from 22% in August 2024 to 40% by November 2025 (RACGP's newsGP). At the same time, at least 80% of GPs surveyed by RACGP said they were not at all, or not very, familiar with specific AI tools (RACGP, Health of the Nation 2025).

Is AI safe in healthcare?

It depends on the tool and how you set it up. In Australia, the Privacy Act and the Australian Privacy Principles apply to any personal information an AI tool handles (OAIC). In New Zealand, the Information Privacy Principles do the same job (NZ Privacy Commissioner). Treat any vendor's "it's secure" line as something to go and check.

What is generative AI used for in healthcare?

Mostly drafting. A consult note for a practitioner to check. A social post or newsletter for someone to edit. A first pass of website copy. A person still makes the decisions.

Do AI scribes require patient consent in Australia?

AHPRA's guidance lists informed consent as one of five duties when using AI in healthcare (AHPRA). RACGP's fact sheet notes that some medical defence organisations recommend written consent (RACGP's AI scribes fact sheet). The Guardian reported that Australia's Privacy Commissioner has found "deficiencies in GPs' implementation of AI scribe consent protocols" (The Guardian). So plan on getting consent, and check what your own medical defence organisation advises.