← DisclosureTrailFree checklist · EU AI Act Article 50 · updated 28 Sep 2026
AI Labelling Checklist for Agencies and Studios
Since 2 August 2026, agencies that publish AI-generated or AI-edited content to EU audiences have labelling duties under Article 50. This is the one-page version: what to label, where the label goes, and what to do when one gets missed. Print it or share it with your team.
In one sentence: if your work shows realistic people, voices, places or events that didn't happen, say it's AI-generated, visibly and at first exposure. If you run a chatbot or voice agent for a client, it must say it's an AI in its first message. And keep a written record that you do this.
1. Does this apply to you?
You create or edit images, video, audio or text with AI tools for clients or your own channels.
Some of that content reaches people in the EU (EU clients, EU audiences, or global campaigns). Where your company is based doesn't matter.
You set up chatbots, voice agents or AI assistants that talk to your client's customers.
Any tick means Article 50 is likely relevant. You are usually the deployer (the one using the AI system professionally). The AI tool vendor is the provider and has its own separate duties, such as machine-readable watermarks.
2. The 5 labelling rules
Deepfakes get a visible disclosure. Realistic image, audio or video content resembling real people, objects, places or events that falsely appears authentic must be disclosed as AI-generated or manipulated (Art. 50(4)). This includes realistic product scenes, synthetic spokespeople and cloned voices.
Creative work gets a lighter touch, not an exemption. For evidently artistic, creative, satirical or fictional work, you still disclose, but in a way that doesn't spoil the work (for example, end-card or credits).
Human review does not remove the label for images, video or audio. The editorial-review exemption applies only to AI-generated text published to inform the public on matters of public interest, with a named person holding editorial responsibility.
Chatbots and voice agents say they're AI first. Unless it's obvious to a reasonable person, people must be told they're interacting with an AI system (Art. 50(1)). As deployer, make sure the client's configuration does this.
Clear, early, accessible. The disclosure must be clear and distinguishable, given at the latest at first interaction or exposure, and accessible (Art. 50(5)). A line in the terms of service doesn't count.
3. Where the label goes, by format
Format
Put the label
Avoid
Video / reels / ads
On screen in the first seconds, readable size, long enough to read
Only in the caption or description
Static social / display
On the image, or the first line of caption before "more"
Hashtag at the end, hidden in alt text only
Audio / podcast / voice clone
Spoken at the start
Show notes only
Chatbot / voice agent
First message or greeting
FAQ page or terms
Website hero / product imagery
Caption or overlay next to the image
Site footer
Artistic / fictional pieces
End card, credits or caption
No label at all
The Commission's Code of Practice introduces a common EU "AI" icon for deployers. Using it alongside a short text label ("AI-generated" / "Made with AI") is the safest pattern for now.
Get this checklist plus a staff one-pager by email
One email with the checklist and a printable staff briefing. Occasional updates on Article 50. Unsubscribe any time.
4. Your internal process
A written AI labelling policy: what gets labelled, which label, who decides.
An AI-use field in your briefing or job sheet, so every asset records whether AI was used and how.
A pre-publish check: someone confirms the label is present before anything goes live.
A short staff briefing so designers, editors and social managers know the rules.
A client clause in your contract or SOW covering who is responsible for labelling on the client's channels.
5. When a label gets missed
Add the label, or take the content down, as soon as it's spotted.
Log it: date, asset, channel, what was missing, who fixed it and when.
Tell the client if it ran on their channels.
Check whether the pre-publish step failed, and fix the process, not just the post.
Key dates
2 August 2026: Article 50 transparency duties apply. The digital omnibus delayed the high-risk rules, not these.
2 December 2026: end of grace for providers' machine-readable marking on systems already on the market. That's a provider duty, not the agency's visible label.
Fines for breaching Article 50 go up to €15 million or 3% of worldwide annual turnover.
The Deployer Pack gives you your AI labelling policy, client contract clause, staff one-pager, pre-publish checklist and correction log, written for your agency. Fixed price, delivered in 48 hours.