How to label AI-generated content: 5 steps under the AI Act
Article 50 of the EU AI Act has applied since 2 August 2026. Find out which AI content you must label, who is responsible and how to set it up in five steps.


Contents
Labelling AI-generated content is a duty under Article 50 of the EU AI Act: people have to know when they are looking at or reading something that AI generated or manipulated. The rule has applied since 2 August 2026, both to companies that build AI tools and to companies that simply use them.
Below are five steps to work through it in your business. Each one has checklist questions, and the sources are at the end. Legal status as of 22 October 2026. This is not legal advice, just a practical map.
Who does Article 50 apply to, and from when?
The AI Act separates two roles. A provider builds an AI system and places it on the market (for example, the maker of an image generator). A deployer uses an AI system in its business. Most small and medium-sized businesses are deployers.
Article 50 puts duties on both. Providers must, among other things, mark generated content in a machine-readable format (Article 50(2)). Deployers must disclose deepfakes and some text (Article 50(4)). The obligations apply from 2 August 2026. One exception: providers of systems placed on the market before that date have until 2 December 2026 to add machine-readable marking. That change came with the Digital Omnibus, Regulation (EU) 2026/1744.
Step 1: List where AI creates content. Where does it happen?
Start with an inventory. You can’t label what you don’t know about. In many companies AI writes social posts, generates ad images, drafts product descriptions, voices videos and answers customers in a chat. Often each team does it its own way.
Checklist:
- Which AI tools are in use (including ones people use on personal accounts)?
- What leaves the company from those tools: text, images, video, audio?
- Who publishes it, and where?
If you can’t answer one of these, ask the team directly. A short survey is enough.
Step 2: Work out your role. Are you a provider or a deployer?
If you use off-the-shelf tools (ChatGPT, Midjourney, a voice generator), you’re a deployer. Machine-readable marking (such as a watermark in the metadata) is then the tool provider’s job, not yours.
That changes when your business builds an AI system and makes it available to others. Take a chat on your website that talks to customers. Article 50(1) requires such a system to be designed so people know they are talking to AI, unless that’s obvious. That’s why, when we build AI assistants, we always add a clear notice at the start of the conversation.
Checklist:
- Do we only use ready-made tools, or do we build something and offer it to others?
- Do customers talk to a chat or an assistant of ours?
- Do they know from the first message that it’s AI?
Step 3: Find the content you must label. What counts as a deepfake?
A deployer has two main duties under Article 50(4).
The first is a deepfake: an image, audio or video generated or manipulated by AI that resembles real people, objects, places or events and could pass as authentic. A generated “CEO” in an ad, or a realistic photo of an office that doesn’t exist, is a deepfake. For clearly artistic, satirical or fictional work the duty is lighter: you disclose it in a way that doesn’t spoil the work.
The second is text published to inform the public on matters of public interest. There’s an important exception here: no label is needed if the text went through human review and a person or company holds editorial responsibility for it.
Checklist:
- Do we show realistic AI-generated images, voices or videos of people or places?
- Do we publish text on public matters (for example, commentary on new rules or information for residents)?
- Does every such text have an editor who checks it and signs it off?
Step 4: Choose how to label. How and when should the label appear?
Article 50(5) says the information must be clear and distinguishable, appear no later than the first interaction or exposure, and meet accessibility requirements. So not in small print in the footer, but where people actually look: on the image, at the start of the video, next to the text.
The Code of Practice on marking and labelling AI-generated content helps in practice. The European Commission published the final version in June 2026. Following the code is voluntary; the Article 50 duty is not. The Commission has also prepared a set of icons you can use for labels.
Checklist:
- Where will people see the label on each type of content (post, ad, video, article)?
- Can screen reader users perceive it too (for example, in alt text)?
- Do we use one label style across the whole company?
Step 5: Write down the rules and train the team. Who looks after this day to day?
The last step turns a one-off clean-up into a habit. Write short rules: which tools you use, what you label, how, and who approves publication. One page is enough. You can see an example on our AI use policy page.
Then walk the people who publish content through the rules. Otherwise they stay in a folder. A short training on using AI safely covers it in an afternoon.
Checklist:
- Are the rules written down and available to everyone who publishes?
- Who approves AI content before it goes live?
- When do we review the rules (for example, after a tool change or new guidance)?
On your own or with help?
If you publish a few posts a month, you can work through this list yourself. When AI content comes from several teams, it’s worth doing it properly once: inventory, rules, training. At our next webinar on 17 November (in Polish) we’ll go through it with examples.
Labelling rules belong in your internal guidelines, as we explain in our guide to a company AI policy.
Summary
Article 50 of the AI Act has applied since 2 August 2026. First, check where AI creates content in your business. Then work out whether you’re only a deployer or also a provider. Label deepfakes and public-interest text that hasn’t been through human review. The label must be clear and visible from the first contact. Finally, write down the rules and train the team.
One thing to do today: ask your team which AI tools they use. That’s step one of five.
Sources
- Regulation (EU) 2024/1689 (AI Act), Article 50, EUR-Lex.
- Code of Practice on marking and labelling of AI-generated content, European Commission.
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), Official Journal of the EU, 24 July 2026.
- AI Act
- Article 50
- AI labelling
- deepfake
Written with the help of AI, edited and checked by a human.



