There’s a small line creeping onto the bottom of more and more content: “Created with AI.” Platforms demand it. Governments are drafting it.
Meta began auto-labelling ad media in June; Google followed in July. The EU is legislating like nobody’s business. Agencies now slap it on everything from a generated film to an ad where someone removed a lamppost.
It’s meant to make audiences feel safer. I think it does the opposite. For advertising, it’s worse than useless, because it asks a question nobody has ever asked: is this ad real?
Let me concede the important ground first. News, evidence and identity should be protected at all costs. A deepfaked prime minister declaring war is not a creative choice, it’s an attack. Synthetic imagery of real events, fabricated documents and cloned voices deserve labelling, watermarking, provenance standards and, sometimes, criminal penalties. Build those systems. Fund them. I’m completely with you.
But somewhere between “we must protect the information ecosystem” and “we must label all content”, the argument lost the plot. It quietly became: audiences deserve to know when what they’re watching isn’t real. Applied to advertising, that principle is incoherent, because advertising has never been real.
The car in the ad has never driven that particularly tricky track. The perfect nuclear family eating breakfast in the kitchen has never existed.
The milk splash is gelatin, the steam is dry ice, the happy dog is with a trainer with treats off camera. Real actors, on real sets, fabricating real realities. CGI has been constructing photoreal worlds for decades. Animation constructs whole ones. Every tool we pick up is invented to build a more persuasive fiction, and audiences have always known it.
Nobody has ever filed a complaint with Ad Standards because the couple in the insurance ad were married to other people. When I was an actor I donʼt think anyone checked my postcode when I pretended to be a Queenslander cracking open a beer I am a hop fiend for anyone still in doubt).
That’s the category error at the heart of blanket AI labelling. It treats synthetic content as attempted deception about reality. But audiences don’t experience a brand film as reality. They experience it as advertising, a format we all agree is constructed, like a sitcom set that isn’t a real apartment. Magritte painted a pipe and wrote “this is not a pipe” nearly a century ago. We are now demanding a government label confirming it.
The second problem is definitional: what exactly are we labelling? The industry has quietly dissolved “AI” into every step of the workflow. Generative fill cleaning up a sky. AI upscaling. AI-assisted grading, de-noising, mixing, translation, scene extensions, retouching, rollout, storyboarding.
Unless you are manually cutting up your analog film, it has touched a tool with AI somewhere in it, I tip my hat to you Mr Nolan). Which means any commercial made with current tools use AI or are augmented by it somewhere between brief and master file. If the label goes on all of that, it goes on everything, and a label that’s on everything tells you nothing.
We’ve seen this: platforms famously slapped “made with AI” warnings on genuine photographs because detection couldn’t tell the difference.
Photographers were furious, and audiences learned the label itself can’t be trusted.
So what should we do instead? Regulate claims & misinformation, not tools.
Australian advertising law already works this way, and on the whole it works. We have issues in political advertising but this is another matter.
The claim must be true. Testimonials must be genuine. The claim is the thing, however the pixels were produced. None of those rules asks which brush you painted with. They ask whether you’re deceiving someone about a fact that matters. Marcel Duchamp famously made the readymade series with ordinary, mass-produced manufactured objects that the artist selected and designated as works of art. To reject traditional beauty and technical skill in favour of visual indifference and ideas.
That’s the test, and it beats “was AI involved?” Did you fabricate evidence? Did you invent a news event? Did you generate a real customer who never existed, or a real scientist who never said that? Then you’re in trouble, whether you generated it or built it in a props department. Did you make a talking moose sell iced coffee? Congratulations, that’s advertising and extremely moosey! The audience was never confused about which part is real, because none of it is, and that’s fine.
Here’s the risk of getting this wrong. Every meaningless “made with AI” badge trains people to read labels less carefully. Alarm fatigue is real: when everything carries a warning, people stop reading warnings, including the ones that protect them. We will have built compliance theatre, where agencies audit the process while the ad beside them lies about the effects of an immigration policy.
For the record, I have skin in this game. My agency runs a purpose built operating system that has AI running through the whole thing, an AI powered creative testing platform and this year 60% of our production has become fully generative. Make no mistake we are 100% red pilled.
If labelling genuinely protected audiences, I’d be first in line. It doesn’t. It creates a fabricated problem and distraction to the real issue of misinformation.
So by all means, protect the places where reality actually hangs in the balance: news, elections, evidence, identity. Give provenance standards teeth there, and let the AANA finish its review with that distinction in mind. But an ad is not a news bulletin, and nobody has ever mistaken one for the other. Save the label for the things that need saving. The rest is just advertising, doing what advertising has always done: being beautifully, transparently unreal.
Nick Hunter is founder CEO & ECD of Paper Moose.
