EU AI Act for Marketers: A Compliance Checklist

On August 2, 2026, the EU AI Act’s transparency obligations took effect, and marketing teams are squarely inside them: Article 50 is the part of the Act written for chatbots, AI-generated content and synthetic media rather than for high-risk systems like credit scoring. (The Act regulates the AI systems themselves - the personal data flowing through them answers to GDPR, which has its own marketer’s guide.) If your team uses generative AI for content, runs a chatbot, or ships AI-made creative into EU-facing channels, some of this now applies to you, with fines of up to €15 million or 3% of worldwide annual turnover behind it.

I am a marketer, not a lawyer, and this is a working translation rather than legal advice. What I can do is what I did for my own work: read the European Commission’s own FAQ on Article 50, sort the five obligations by whether a marketing team is actually the party on the hook, and turn the result into a checklist you can hand to whoever owns compliance.

First, which hat you’re wearing

The Act splits duties between providers, who build or commission AI systems and put them on the EU market, and deployers, who use AI professionally. A marketing team using ChatGPT, Midjourney or a vendor chatbot is a deployer. You are only a provider if your company ships an AI system of its own - and then a different, heavier set of duties applies that this post doesn’t cover.

That split does most of the reassuring work, and it’s the first thing I check when a new rule lands. Two of the five obligations, including the technically hardest one, belong to your vendors, not to you.

The five obligations, sorted for a marketing team

1. Chatbot disclosure - your vendor’s duty, your surface. Providers of AI systems that interact directly with people must make sure users know they’re talking to AI, unless it’s obvious from context, from the start of the first interaction. The duty sits with the provider, but the widget sits on your website with your logo above it. My rule is to label it anyway: “AI assistant” in the chat header costs nothing and removes the argument entirely.

2. Machine-readable marking of AI output - your vendor’s duty, worth verifying. Providers of generative AI must mark synthetic audio, image, video and text in a machine-readable format so it can be detected as AI-generated. This is watermarking at the model level, and it is OpenAI’s, Google’s and Adobe’s problem to build. Your job is narrower: know whether the tools in your stack comply, and stop routing output through steps that strip the marking.

3. Deepfake labelling - yours. Deployers must clearly disclose content that appreciably resembles real people, places or events and would falsely appear authentic. In marketing terms: an AI avatar of your founder, a cloned voice reading your ad, a photoreal scene that never happened. Artistic and satirical work gets a lighter “appropriate” disclosure, but ads have no artistic exemption I would want to test.

4. Labelling AI-written text that informs the public - yours, with the exemption that saves content marketing. Deployers must label AI-generated text published to inform the public on matters of public interest. Most product and brand copy is not that. Your thought-leadership piece on public health, financial wellbeing or the environment might be. The exemption that matters: the label is not required where the content has undergone human review and a person holds editorial responsibility for it. A real editorial pass - the kind a decent content operation runs anyway - is now also a compliance control, and worth documenting for that reason.

5. Emotion recognition disclosure - rarely yours, watch your martech. Deployers of emotion recognition or biometric categorisation systems must inform the people exposed to them. Most marketing teams run nothing of the sort. Some personalisation and ad-tech vendors flirt with it, which is a question to put in your next vendor review rather than a reason to panic.

The checklist

What I would actually run through this month, in order of effort:

1. Inventory every AI touchpoint that faces EU users: chatbots, AI-generated blog and social content, synthetic voices and avatars, AI imagery in ads, personalisation systems. You cannot label what you haven’t listed.
2. Label the chatbot in its header or first message. Five minutes.
3. Add a disclosure line to synthetic media - AI avatars, cloned voices, photoreal generated scenes - in EU-facing campaigns.
4. Document your editorial review. For AI-assisted articles, record who reviewed what and who holds editorial responsibility. This is the difference between needing a label on public-interest content and not.
5. Ask your genAI vendors the marking question: how do your outputs carry machine-readable AI marking, and what breaks it? Keep the answers.
6. Check your ad platforms’ own rules, which now run ahead of the law. Google added “How this ad was made” disclosures across Search, YouTube and Discover in July 2026 - automatic when you use Google’s AI ad tools, self-declared when you bring AI-made creative from elsewhere. Declare honestly, because the platform policy and the regulation now point the same way.
7. Diarise December 2, 2026. Systems already on the market before August 2 got a grace period on the marking obligation only, and it ends then. Content generated before August 2 does not need retroactive labels.

What I’d not do

Two overreactions cost more than compliance does. Slapping “made with AI” on every asset your team touches is one: the Act requires disclosure in specific situations, and labelling everything trains your audience to read the label as noise while implying your editorial process does less than it does - the disclosure question gets my full answer separately, trust research included. Banning generative AI outright is the other, and it usually just drives usage underground where none of the checklist above happens. The Act, read plainly, rewards the boring thing: knowing what you use, reviewing what you publish, and saying what’s synthetic when a reasonable person would want to know.

This sits inside a bigger picture - the US half of the regulatory map is in my read of the FTC’s AI advertising rules, the manipulation tactics European authorities sweep retail sites for are in the dark patterns guide, budget owners will find the governance framing in the CMO playbook, and the editorial-review step is the same human pass that content marketers should be running for quality regardless of what Brussels thinks. The ethics and governance guides will grow from here, and the wider context lives in the complete guide to AI marketing.

One more time, because it matters: this is my working read of the Commission’s FAQ as a practitioner, current as of early September 2026. Anything with real money or real risk attached goes to your lawyer, with the inventory from step one in hand.