As of August 2, 2026, the EU AI Act’s transparency obligations are in force. The European Commission adopted its guidance on Article 50 on July 20, and the deadline arrived without postponement.
Most non-European marketers have filed this under «European problem.» That’s the expensive misreading. The obligations follow where the content is shown, not where the company is registered. An agency in Dubai running a campaign to an audience in Germany is inside scope. So is a creator in Kyiv whose content reaches a diaspora audience in Poland.
Here’s what has to be disclosed, what doesn’t, and where the line actually falls.
A note before we start: this is a summary of publicly available guidance, not legal advice. For a specific campaign, a specific market or a borderline asset, talk to a lawyer who works in this area. The stakes described below are the reason.
What Article 50 requires
Four disclosure categories:
1. AI interaction. If a person is interacting with an AI system, they must be told. A chatbot has to identify itself as a machine rather than let a user assume they’re speaking to staff.
2. AI-generated content. Synthetic images, audio and video must be disclosed as artificially generated.
3. Deepfakes. Content depicting real people or real events that didn’t happen carries its own disclosure requirement.
4. AI-generated text on matters of public interest. Where AI-written text addresses topics of public significance, that has to be disclosed too.
The penalties: up to €15 million or 3% of global turnover. For an agency, that’s not a compliance line item. That’s the business.
The territorial rule that catches people
This is the part worth reading twice.
The obligation attaches to where the advertising is shown, not where you are. Your company’s registration, your bank, your team’s location — none of it exempts you if the audience is in the EU.
Practical situations that are in scope and rarely recognised as such:
- A campaign targeting a diaspora audience that happens to sit in EU member states
- Broad social targeting that isn’t geofenced away from Europe
- Organic content on a platform with EU reach
- A client whose customer base extends into the EU even though the business doesn’t
If you have never explicitly excluded the EU from a campaign’s delivery, assume you’re covered.
Where the line sits: retouching versus creating
The most useful distinction in the guidance, and the one people get wrong in both directions.
Ordinary photo retouching does not require labelling. Colour correction, skin smoothing, removing a stray object, adjusting exposure — standard post-production doesn’t trigger disclosure.
Creating something realistic that never existed does. A photorealistic product shot generated from a text prompt. A model who doesn’t exist wearing your clothing. A location that was never photographed. A person saying words they never said.
The test is roughly: did you improve a real thing, or manufacture a plausible unreal one?
The awkward middle ground is real. Generative fill that extends a genuine photograph’s background. A real product placed into a synthetic environment. An authentic photo with an AI-generated element added. These are exactly the cases where the guidance stops answering cleanly and a lawyer starts earning their fee.
What Digital Omnibus did — and didn’t — change
There was a widespread assumption that this deadline would slip. It didn’t.
The Digital Omnibus (Regulation (EU) 2026/1744) entered into force on July 27, 2026 and postponed the AI Act’s high-risk obligations to 2027–2028. That’s a genuine delay, and it generated headlines suggesting the whole framework had been pushed back.
The August 2 transparency deadline was not moved. If your understanding of the timeline came from Omnibus coverage, check it against the actual scope of that postponement.
What this means asset by asset
AI-generated ad visuals. Disclosure required where the image depicts something realistic that wasn’t photographed. Note that Google’s image and video models embed SynthID watermarking by default — a useful provenance record, though not a substitute for a visible disclosure.
AI avatars and synthetic presenters. Squarely in scope. A generated spokesperson delivering your message is a synthetic depiction of a person.
Customer service chatbots. Must identify themselves as AI. This is the easiest requirement to satisfy and the most commonly overlooked, usually because a brand deliberately designed the bot to feel human.
AI-assisted copy. Ordinary marketing copy drafted with AI assistance is not what the public-interest text provision targets. Content addressing matters of public significance is a different question — and if you’re publishing in that territory, get specific advice.
Retouched product photography. No disclosure required for standard post-production on real photographs.
The strategic layer underneath the compliance one
Here’s what makes this more than a legal chore.
Consumer research points in an uncomfortable direction for anyone planning to comply by labelling heavily. Harris Poll findings presented at Cannes Lions showed 63% of consumers less likely to buy from a brand using AI-generated advertising and 73% less likely to trust an ad they suspect was AI-made. Klaviyo/Datalily research across 8,000 consumers in eight countries found only 7% say visible AI content increases brand trust, while 31% say it decreases it.
So the regulation forces you to disclose exactly the thing your audience penalises you for.
Which reframes the whole question. The efficient response isn’t a better labelling system. It’s using less synthetic content in the places customers actually look — and moving AI into the parts of the operation that were never customer-facing: research, variant generation, drafting, analysis, internal systems.
Compliance cost and audience preference are pointing the same way. That’s rare, and it’s a strong signal about where the line between AI-assisted and AI-made should sit in your work.
A practical checklist
1. Map your EU exposure. Which campaigns, which clients, which organic channels reach audiences in the EU? Most teams have never asked.
2. Inventory your synthetic assets. Everything realistic that was generated rather than captured. Split it into disclose / doesn’t apply / ask a lawyer.
3. Check every chatbot. Does it identify itself as AI on first contact? This is a same-day fix.
4. Decide your disclosure format now. Not per-campaign improvisation — one consistent approach, agreed with legal, applied everywhere.
5. Write it into client agreements. If you produce content for clients, define who is responsible for compliance in each market. Do it before an issue arises, not during one.
6. Get advice on the borderline cases. Composites, extended backgrounds, real products in generated scenes. These are genuinely unclear and the penalty range makes guessing a poor strategy.
About the author
Alina Palii — brand strategist, founder of ALPA Marketing.
She works with founders and leadership teams on the decisions that come before the marketing: what the brand stands for, who it is genuinely for, what it declines to be, and how that translates into everything the market eventually sees. Strategy first — the content, the channels and the campaigns follow from it.
10+ years in marketing and a master’s degree in the field. She has built brands from zero for AI startups and national companies, and shaped the positioning of personal brands whose audiences buy on trust rather than on price.
Ukrainian by origin, living between Dubai, Paris and Ukraine, and working across the UAE and European markets — a vantage point that matters when a brand has to hold its meaning across cultures rather than be rebuilt in each new one.
She works across categories rather than inside one. Positioning logic travels between industries even when the audience doesn’t, and the pattern recognition that comes from moving between them is often what a category-blind competitor is missing.
Alina takes on a limited number of strategy engagements at a time.
