We were speaking at ADX earlier this year, running through the advertising guidelines the way we usually do. Before and after images have to be true. They have to be the patient’s own. You have to have consent to use them.
A few people in the audience turned out to be AI software developers, and one of them asked a question that stopped the room.
So are you saying you cannot use AI generated images for those before and afters?
Yes. That is exactly what we are saying.
They were building a tool that generated them.
The rules have not changed. The tools have
There is nothing new in the requirement itself. AHPRA’s advertising guidelines have long been clear that testimonials, claims and images used to advertise a regulated health service must be accurate, must not be misleading, and must not create an unreasonable expectation of benefit. Before and after photography sits squarely inside that. The image has to show a real outcome for a real patient, and you need that patient’s consent to use it.
What has changed is that producing a persuasive, entirely fictional before and after now takes about ninety seconds and no photographic skill at all.
That gap is the whole problem. The obligation is unchanged and the temptation has become enormous, particularly for cosmetic work where the image does most of the selling.
To be clear about what we could and could not find: we are not aware of an AHPRA publication addressing AI generated imagery specifically. What exists are the general advertising requirements, which do not need a special AI clause to apply. An image that was never a patient cannot be a true representation of that patient’s outcome, whatever produced it.
Why the practice carries this, not the software company
This is the part practice owners consistently get wrong, and it is the reason we keep raising it at every event we speak at.
If an AI tool produces marketing that breaches the advertising guidelines, the person in breach is the registered practitioner and the practice. Not the company that built the tool. Not the agency that ran the campaign. You.
Sean Perera, Chief Technology Officer at Centaur Software, made the same point from the software side when we spoke on the AI Your Practice podcast. His concern was how the industry stops tools reaching the market that cannot comply from the concept stage. The practice that uses one is trusting an AI company to turn up and deliver, and finding out later that the product was in breach before it was written.
Buying from a vendor does not transfer a professional obligation. There is no version of this where the software company answers the notification.
The four situations where this actually bites
In practice, this is not usually a case of somebody deciding to fabricate results. It is more mundane than that, which is what makes it easy to walk into.
A tool that generates the image outright. The clearest breach and the easiest to spot. If the picture is of a face that does not exist, it cannot be a true before and after.
A tool that enhances a real one. Harder, and much more common. Whitening the teeth a little, evening the lighting, smoothing the skin, removing the shadow that made the before look worse than it was. The patient is real, the outcome is real, and the image is now misleading. The consent you hold does not cover an image that was altered after they gave it.
A stock or illustrative image used as an example. If it sits next to your results and reads as your work, the caption underneath it will not save you.
A tool that generates a simulated outcome for a patient consultation. This one is clinically useful, and the risk is what happens next. A simulation created for a chairside conversation is not marketing material, and it should never end up on the website or an ad account.
What to actually do about it
Five things, and none of them are difficult.
- Ask any marketing tool, plainly, whether it generates or modifies images. Get it in writing. Describing itself as AI powered tells you nothing about which part is AI powered.
- Separate simulation from marketing at the point the image is created. If a tool produces predictive outcome images for consultations, keep those in a place your marketing never draws from.
- Check your existing consent forms cover the image, not just the treatment. Consent to be photographed is not consent to appear in advertising, and neither covers a version of the photograph that has been altered.
- Audit what is already published. Go through the website and the social accounts and find who produced each before and after and when. Most practices have at least one image nobody can now account for.
- Write down who signs off marketing images. If the answer is whoever posts that day, that is the actual risk, and it costs nothing to fix.
The bigger pattern this sits inside
The developers at ADX were not acting in bad faith. They had built something clever, in a market moving quickly, without anyone in the room who knew the advertising guidelines existed.
Sean expects that to produce consequences before it produces caution. His comparison is the early years of driverless cars, where the technology arrived before the case law did, and the pattern resolved through a small number of very public incidents. He anticipates the same in dentistry: a few significant breaches, some landmark rulings, and practices facing serious claims over patient privacy.
We think he is right, and the uncomfortable part is that the first practices to find out will not be the ones that were careless. They will be the ones that trusted a supplier and never asked what the tool was doing.
There is a related question worth asking while you are here, which is whether your insurance actually covers any of this. Most owners have not checked.
What to do this week
Open your website and count your before and after images.
For each one, answer two questions. Who took it, and has it been altered since. If you cannot answer both for any image, take that image down until you can.
That is the entire task, and for most practices it takes half an hour. It is also the single cheapest compliance action available to you right now, because it happens before anyone else has to ask.
Watch it again
The full conversation is on YouTube: watch it here.
Related reading: the five checks before you sign with any dental AI vendor.
The practice that has used this tool are going to be the ones in breach of AHPRA guidelines.
Carolyn S Dean
From the AI Your Practice podcast with Sean Perera, Chief Technology Officer at Centaur Software.
This article is written around the existing requirement that advertising images be true and patient consented. It does not claim AHPRA has published guidance specifically on AI generated images.
