Not every ad touched by Midjourney needs an AI label. Not every line drafted in ChatGPT does either. That is the core message from VIA Nederland, the Dutch advertising industry association, in new guidance aimed directly at creative agencies.
The association walks agencies through Article 50 of the EU AI Act — the transparency chapter that became legally applicable across the European Union on August 2, 2026. Instead of restating the regulation, VIA Nederland built its guidance around four everyday agency activities and asked a simple question for each: does a disclosure duty attach, and what form does it take?
First, know your role: provider or deployer
Everything hinges on one distinction. A provider develops an AI system (or has one developed) and puts it on the market under its own name. A deployer uses an AI system under its own authority for professional purposes.
According to VIA Nederland, most creative agencies are deployers — they use ChatGPT, Midjourney or Adobe Firefly rather than building models. The line blurs when an agency builds a custom application on top of an existing model. At that point, each application may need its own assessment.
Why it matters: providers face marking duties designed to make output machine-detectable. Deployers face labeling duties designed to make it human-detectable. Different obligations, different workflows.
The four triggers
- Chatbots and brand assistants. When an AI system talks directly to people, users must know they are talking to AI. The notice has to be clear at first contact — a short on-screen line for a chatbot, a spoken one for a voice assistant. Burying it in terms and conditions does not count. An exception applies where it is already obvious to the user, and context decides what “obvious” means.
- Image, audio and video. The test is whether the finished asset qualifies as a deepfake: AI-generated or manipulated content resembling real people, places, objects or events in a way that could wrongly seem authentic. Notably, an AI-generated person does not need to resemble a specific celebrity — it is enough that they could be mistaken for a real person. A commercial with realistic AI actors is in scope. Colour grading tweaks and clearly fictional VFX are not.
- Text. Much narrower than agencies fear. The duty applies only to text published to inform the public on matters of general interest — politics, public health, public safety, fundamental rights. Routine ad copy is not covered. And an exception applies where genuine human editorial review takes place. A spell-check does not qualify; someone with relevant knowledge must be able to assess, adjust or reject the content, and VIA Nederland recommends documenting it.
- Building your own AI tool. Rare, but consequential. Build a generative layer on an existing model and you may become a provider for that tool — which means considering machine-readable marking of your output, independent of whatever the base model already does.
The artistic exemption will not save your campaign
Here is the part agencies will want to misread. The AI Act carves out creative, artistic, satirical and fictional works — but only to allow a less prominent label, not to remove the duty. And where the commercial message dominates, that lighter regime cannot be invoked at all. VIA Nederland states plainly that the exception will not apply to creative advertising messages.
On mechanics: metadata alone is not enough. An invisible watermark alone is not enough. The person seeing or hearing the content must actually be able to perceive the disclosure — text, visual or audio. The Commission’s icon set is optional; you can design your own.
The commercial catch
Compliance and performance do not automatically point the same way. IAB research published January 16, 2026 found 82% of ad executives believed Gen Z and Millennial consumers felt positively about AI-generated ads — against just 45% of those consumers who actually did. That gap widened from 32 points in 2024 to 37 points in 2026. A separate Raptive study found suspected AI content cuts reader trust by close to half, with a 14% drop in purchase consideration.
Translation: blanket-labeling everything is not the safe option. Over-labeling costs trust and click-through. Under-labeling carries legal exposure — and the EU AI Office has already opened three complaint routes covering transparency breaches.
What to do Monday
VIA Nederland’s advice is to build transparency into the creative process, not bolt it on before launch. Run five checks per campaign: define your role; identify the specific application; test image, audio and video for deepfake risk; document substantive human review; then decide if and how you disclose.
Practical version for agency teams: add a “disclosure” row to your creative brief template. If the concept involves realistic synthetic humans or cloned voices, design the label into the layout at storyboard stage — because you cannot claim the lighter artistic regime later.
Source: PPC Land
