Wondering whether an AI writing tool can be trusted with advertising that a regulator might one day read? Curious what an AHPRA advertising breach actually costs a practice, and whether that is one penalty or a page full of them?
In this article you will find why Anthony Middlemiss walked away from allied health after twenty years, which two jobs pushed him out, why marketing behaves like an appetite rather than a task, how compliance can be written into a product rather than bolted on afterwards, and what an AHPRA breach really costs when somebody goes looking.
This article was created together by Anthony Middlemiss and Carolyn S Dean. For more about Anthony, scroll to the end of this article.
The founder who was the customer first
Anthony Middlemiss did not arrive at practice software through technology. He arrived through twenty years of running allied health practices, and through the decision to stop.
His first career was network engineering. He retrained at a college that taught allied health, spent about two decades in practice, and then left. The detail that matters is what he left because of. It was not the clinical work. It was everything that came bolted to it: the marketing, the staffing, the HR and the compliance.
“You wanna be helping people, not directing people,” he tells Carolyn, and that single line explains the whole arc. Nobody enters a health profession because they want to run a roster or work out what a practice is legally allowed to say about itself. Yet that is what ownership becomes, and almost nobody is trained for it before the day it lands on them.
So he went back to IT, took a post graduate course in AI and machine learning, and got close enough to the technology to form a view on it. He describes seeing both sides: how capable it can be, and how unsettling it can be where the rules are strict and the consequences are personal. That combination, the practice years and the retraining, produced Trilbii AI. He built the product that would have solved his own problem.
Carolyn accepts the premise and extends it. The people best placed to fix practice software, she argues, are the ones who left practice because of it. Everything Anthony says after this rests on that ordering: practice first, technology second.
HR and marketing pushed him out of practice
Carolyn asks him directly which parts of running a practice he disliked most. The answer comes back without hedging.
“For me, it was HR and marketing.”
HR is a memory problem, not a people problem. Anthony’s description of HR is oddly precise. It is, in his words, dealing with people and their inability to remember things, and maintaining checks and balances that are not always up to date. That is not a complaint about staff. It is an administrative burden that never quite resolves, because the documentation drifts out of date faster than anyone can keep it current.
Marketing is laborious, boring, and expensive to hand over. His words: “Marketing is, like, so laborious and boring, and so expensive to get someone else to do.” That sentence contains the trap most practice owners are sitting in. Doing it yourself costs time you do not have. Handing it to an agency costs money that is difficult to justify against results you cannot easily measure. Neither option feels correct, so most owners oscillate between them for years.
Carolyn recognises both from the practices she works with, and she is candid that the second one is the reason her agency exists at all. Owners are not avoiding marketing because they are uninterested. They are avoiding it because both available routes are unattractive, and no third option has existed.
Marketing is not a task. It is an appetite
This is where the conversation produces its only metaphor worth keeping, and both speakers reach for it independently.
The problem is the overwhelm, not the difficulty. Anthony is careful here, and the care is what makes it credible. He does not claim to have hated marketing. “It’s the overwhelm,” he says. “‘Cause you get to go through periods where you enjoy it, but it just needs to be fed constantly.” Some weeks he liked it. What broke him was that it never finished. A task you dislike can be scheduled. An appetite cannot.
The beast is getting bigger. “That beast is getting bigger and bigger and needs to be fed more and more.” The volume of content a practice is expected to produce is not levelling off. More channels, more frequency, more formats, and now search itself is changing shape. The manual approach does not just stay hard, it gets steadily worse, and the gap between what a practice should publish and what it actually publishes widens every year.
Time, understanding and inclination rarely arrive together. Carolyn extends the point from what she sees across her client base. Practices know they need to feed the beast, she says, and they do not have the time, or the understanding, or the inclination to keep going. Any one of those three missing is enough to stall a marketing programme, and most practices are missing at least two.
That reframes what a tool like this is actually for. It is not replacing a marketer. It is relief from an appetite that no owner can satisfy on their own.
Built in, not bolted on afterwards
This is the technical middle of the episode, and it is where Anthony stops being a former practice owner and starts being a builder. He describes three decisions and, more usefully, the reasoning behind each.
The rules live inside the program. AHPRA’s advertising requirements are not something the system goes and fetches when it needs them. They sit inside the product, which means the model is working within a constraint rather than consulting a reference and forming its own interpretation. His reason is blunt: “We can’t afford hallucinations, um, where it just makes stuff up because it thinks it’s compliant.” A general purpose chatbot has no way of knowing what a regulated health profession cannot say. It will produce something confident and wrong, and confidence is exactly the wrong quality here.
Every piece of content is checked three times. The system prompts, then checks the output, then rejects and regenerates until the result passes. Carolyn’s response is the most human moment in the episode. “I didn’t actually know that,” she says, on air, about her own product. It is a small thing, but it removes any sense that this was rehearsed, and it shows the same split between a marketing founder and a technical founder that practice owners will recognise from their own businesses.
Slow is the point. Generation can take thirty seconds to a minute, and Anthony does not present that as a limitation to apologise for. It is what the checking costs. If a tool produces compliant content instantly, the reasonable question is what it checked and when, because checking takes time.
What sits underneath all three is his sense of the stakes. “It’s people’s livelihoods. It’s our livelihood. It’s the practice’s livelihoods.” The failure mode here is not an embarrassing sentence. It is a penalty against someone’s ability to keep working.
Pro Tip: Before you use any AI tool to write practice marketing, ask one question: where do its compliance rules come from, and are they inside the system or being looked up? Only a tool with the rules inside it can refuse to produce something.
One breach is not one fine. It is page after page of them
The most commercially useful stretch of the episode is the one where a vague fear becomes a number.
What is the maximum penalty? Up to $60,000 per offence for an individual, and up to $120,000 per offence for a body corporate, for unlawful advertising of a regulated health service. Whether your practice is a sole trader or a company changes which figure applies to you, and most owners have never checked which one they are.
Is it counted per website or per breach? Anthony raises the figure, and Carolyn corrects him on this in real time. It is not one penalty per website. “I believe it’s per breach,” she says. “So when I’ve seen the AHPRA breaches come through, they are literally page upon page upon page of breaches.” That correction is the most valuable thing in the episode, because a practice imagining one fine is mispricing its risk by a wide margin. One investigation does not produce one finding. It produces a report on everything examined, across the website, the Google Business Profile, every social account, email templates and printed material.
Is AHPRA starting to check automatically? Handle this one carefully, because the episode does. Carolyn says she saw a report this week from an organisation she trusts, indicating that AHPRA has AI checks running on the clinical side of cosmetic dentistry. She is explicit that she does not know whether that extends to advertising, and that caution is worth preserving. Anthony’s expectation is that it will: “AHPRA’s probably not too far away from being able to web scrape itself.” That is his expectation about where the direction of travel leads, not a confirmed regulatory position, and it should be read that way.
Where does our data sit, and what is the model pointed at? Carolyn says this is the question practices ask her most, and it is really two questions that vendors habitually answer as one. Can another practice see our data, and will our data train the models? They are separate protections. On the first, each practice sits in its own tenancy in the database, which is what stops data crossing between practices. On the second, Anthony’s answer is architectural: “With the LLMs, we’re pointing the LLMs at the content creation, not at their data.” Practice data is not sent to the model at all, which is a stronger position than a promise about how it would be handled if it were.
On location, Carolyn states the rule plainly: “No patient data and no system that ever touches patient data can live anywhere but in Australia.” Anthony confirms the implementation, being Australian servers, currently two in Sydney, with Perth and Melbourne named as later additions.
All of this traces back to a decision made before the first feature existed. Compliance was written into the code base first, because retrofitting it later is tech debt, and tech debt in a large code base only becomes more expensive. “So the safest bet, I think, especially in this day and age, is to start with compliance first,” Anthony says. He is equally plain about the cost: “So I started it that way, which meant we were slower to kick off.” That delay bought two rounds of validation with real practices before launch, and he would make the same call again.
About Anthony Middlemiss
Anthony Middlemiss is co founder and Head of Customer Success at Trilbii AI, an Australian AI platform built for the non clinical side of dental and allied health practice. He spent about twenty years running allied health practices before leaving the profession, began his working life in network engineering, and returned to technology through post graduate study in AI and machine learning. Before Trilbii AI he built websites for dental practices and repaired AHPRA advertising breaches on them by hand, which is where the compliance first argument came from. Find Anthony on LinkedIn.
Other notes from this episode
- Recorded for AI Your Practice with Carolyn S Dean. The episode runs 14 minutes 40 seconds.
- The penalty figures quoted in this article come from AHPRA’s own guidance. See Advertising and the law. Both figures are maximums and both are per offence.
- The claims about AHPRA using AI monitoring are reported second hand in this episode and are not confirmed by the regulator. Treat them as direction of travel rather than as settled fact.
- A longer article on what an AHPRA advertising breach actually costs follows later in this series and links back to this page.
- Product descriptions here, being the triple check, the tenancy isolation and the Australian hosting, are Trilbii AI’s own account of how it is built. They are not claims about how other tools work.
- Carolyn S Dean is on LinkedIn, where she writes about AI, marketing and compliance for Australian practices.
- Trilbii AI is Australian hosted, with two servers in Sydney, and is built against AHPRA’s advertising requirements. Other jurisdictions, including HIPAA and the EU, are treated as configurations of the same system rather than as separate products.
Watch or listen to the episode
This episode of AI Your Practice, hosted by Carolyn S Dean, features Anthony Middlemiss of Trilbii AI on compliance, data sovereignty and the parts of practice ownership nobody trains you for. Watch it on YouTube.
If the episode was useful, subscribing on YouTube is the thing that helps most, and it takes a second.
Related reading: what an AHPRA advertising breach actually costs.
You wanna be helping people, not directing people.
Anthony Middlemiss, Trilbii AI
From the AI Your Practice podcast with Carolyn S Dean.
