Facts first: Since August 2, 2026, the first transparency obligations under Article 50 of the EU AI Act apply. They affect companies, freelancers and public authorities that use certain AI systems or publish AI-generated content. One exception applies to the machine-readable labeling of AI-generated content: for existing AI systems, this obligation only takes effect from December 2, 2026.
That is why many companies are currently asking themselves the same question: Will we have to label our AI-generated content from now on?
The short answer is: No, not automatically.
The EU AI Act distinguishes between different transparency obligations. Whether labeling is required always depends on the specific use case. Among other things, what matters is what content is published, how artificial intelligence was used, and whether a human reviewed the content and took responsibility for it before publication.
In this article, we use typical situations from marketing, social media and corporate communications to show when labeling may be required and when it is not.
Would you first like to understand which transparency obligations the EU AI Act provides for in general? Then we recommend starting with our article “AI Labeling Obligation: What Applies Under the EU AI Act”.
Why practical examples are often more helpful than legal texts
The EU AI Act deliberately sets out general rules. After all, a law cannot cover every single situation in everyday business. That is exactly why so many questions are coming up right now.
Does a LinkedIn post written with AI need to be labeled? What about a blog article? What applies to an AI-generated product image or a chatbot on your own website?
The following examples are no substitute for legal advice. However, they help you better understand the transparency obligations and assess typical situations in everyday business.
AI-generated texts
Does a social media post written with AI need to be labeled?
➡️ In most cases, no.
Many companies now use artificial intelligence to develop ideas or create first drafts of social media posts. That alone, however, does not automatically trigger a labeling obligation.
A typical post on LinkedIn or Instagram usually serves marketing or communication purposes and generally does not count as a matter of public interest within the meaning of the EU AI Act. On top of that, posts are often reviewed, adjusted and approved by an employee before they are published. It is precisely this human oversight that plays an important role.
Does a blog article written with AI need to be labeled?
➡️ It depends.
Today, blog articles are often created together with artificial intelligence. AI helps with brainstorming, develops outlines or drafts first versions of the text.
Whether labeling is required, however, depends on several factors. Among other things, it matters whether the article deals with a matter of public interest and whether the content was editorially reviewed before publication.
A blog article about your own services or expertise from everyday business is often assessed differently than automatically published content on political or social issues.
Does an AI-generated product description need to be labeled?
➡️ No.
Product descriptions, landing pages or classic advertising copy generally do not fall under the transparency obligation for AI-generated texts under Article 50 of the EU AI Act.
That does not mean, however, that artificially generated content should be published without review. Product information must still be accurate and must not contain misleading statements.
Does a press release written with AI need to be labeled?
➡️ It depends on the individual case.
Press releases can address topics that are relevant to forming public opinion. That is why you should check particularly carefully whether editorial oversight took place and who takes responsibility for the published content.
The more thoroughly a human reviews, revises and approves the text, the more likely it is that the labeling obligation does not apply.
Practical tip: Don't rely exclusively on AI for press releases. Use it to help with the wording, but always do the subject-matter review yourself.
Does a newsletter written with AI need to be labeled?
➡️ In most cases, no.
Newsletters usually serve customer communication or marketing and therefore generally do not fall under the transparency obligation for AI-generated texts under the EU AI Act. Here too, what matters is that a human reviews and approves the content before it is sent.
Things are different for AI-generated images or videos within the newsletter. These should be considered separately from the text. Depending on the subject and how realistic they look, for example with photorealistic depictions or deepfakes, separate transparency obligations may apply.
Do internal documents or emails need to be labeled?
➡️ No.
Internal minutes, emails or working documents are generally not published and therefore usually do not fall under the transparency obligations for AI-generated texts.
Nevertheless, it pays to be transparent about artificial intelligence internally as well and to raise your employees' awareness accordingly.
Practical tip: Define clear company-wide rules for the use of AI. This creates a shared understanding and makes it easier to use artificial intelligence responsibly.
AI images and videos
Does an AI-generated product image need to be labeled?
➡️ This should be checked carefully.
AI-generated product images are being used more and more, especially in online marketing. They are often used to show products in an appealing setting or to replace elaborate photo shoots. However, caution is advised here.
If it is a photorealistic depiction of a real product, a transparency obligation may apply under certain circumstances. The EU AI Act takes a particularly critical view of content that viewers could perceive as real photographs.
Things are different for obviously fictional products or stylized illustrations. Here, there is generally no risk of them being mistaken for reality.
Practical tip: If you publish photorealistic AI images of real products, you should check whether labeling makes sense or is required. When in doubt, transparency builds more trust than uncertainty.
Does a photo edited with AI need to be labeled?
➡️ It may be required.
It is not only fully AI-generated images that can fall under the transparency obligations. Existing photos can now also be altered in just a few clicks. A new background, additional objects or removed people often look deceptively real.
It is exactly these kinds of changes that can fall under the deepfake rules of the EU AI Act. Normal image editing, such as adjusting brightness, contrast or colors, is generally not affected.
Practical tip: Always ask yourself: Could a viewer mistake this image for an unaltered original photo? If the answer is "Yes", you should take a closer look at the use case.
Does a deepfake need to be labeled?
➡️ Yes.
Deepfakes are among the best-known transparency obligations of the EU AI Act. This is by no means only about manipulated videos of celebrities. Images, videos or audio files that were created or altered with AI and look deceptively similar to real people, places or objects can also fall under this rule.
The aim of the rule is to make manipulation transparent to users.
Practical tip: If AI alters content so that it looks like a real recording, you should generally treat it as a potential deepfake and check it accordingly.
Does an AI-generated employee photo or application photo need to be labeled?
➡️ This should be checked carefully.
More and more companies are using artificial intelligence to enhance employee photos or create entirely new images for recruiting campaigns. If these are photorealistic depictions of real people, the transparency obligations of the EU AI Act can become relevant.
Especially if viewers could assume that these are real photographs, you should check whether labeling is required.
Does satirical or artistic AI content need to be labeled?
➡️ Yes – but less prominently.
Art and satire are also taken into account by the EU AI Act. The transparency obligation does not disappear entirely. However, the label may be less prominent, for example in the video description or the end credits.
Humorous content in particular should therefore not automatically be seen as an exception.
Practical tip: A short transparency notice is often enough and at the same time ensures you use AI responsibly.
AI in customer service
Does an AI chatbot on a company website need to be labeled?
➡️ Yes.
If you use an AI chatbot on your website, you must inform users that they are communicating with an AI system. The notice must be given at the start of the conversation at the latest and must be clearly understandable. There is no legally prescribed wording.
A simple notice such as “You are chatting with our AI assistant.” is enough in many cases.
Special cases
Does your own AI system need to label content in a machine-readable way?
➡️ Yes.
This transparency obligation primarily affects companies that develop their own AI systems or offer them under their own name.
This is not about visible notices for users, but about technical labels, for example via metadata or digital watermarks, so that other systems can recognize AI-generated content.
For companies that merely use existing AI tools such as ChatGPT or Claude, this obligation usually plays no direct role.
Practical tip: If you develop your own AI applications, you should look into the technical requirements early on.
Does older AI content need to be labeled retroactively?
➡️ Generally no – but check again if you republish it.
Content published before August 2, 2026 generally does not need to be labeled retroactively.
However, if the same content is published again later or, for example, reposted as a new social media post, you should check the transparency obligations again.
What does this mean for companies?
Above all, the practical examples show one thing: It is not the use of artificial intelligence that determines whether labeling is required, but the specific use case.
Just as important, however, is another idea behind the EU AI Act: humans remain responsible. This approach is often referred to as Human in the Loop.
What this means is that artificial intelligence supports people but does not make decisions on its own. Content is reviewed, assessed and approved before publication. It is exactly this human oversight that also plays a decisive role in the transparency obligations.
Companies should therefore not see AI as a replacement for employees, but as a tool that speeds up routine tasks and supports creative processes.
What does this mean for Matelify?
Matelify follows exactly this idea.
Our Mates support companies throughout the entire social media process. The Mates take on many time-consuming tasks, but responsibility always remains with humans.
Every post must be reviewed, adjusted and approved before publication. Companies stay in control of content, tone and messaging at all times.
In the context of the EU AI Act in particular, this approach is becoming even more important. A clearly defined human-in-the-loop process helps companies ensure quality, create transparent workflows and use artificial intelligence responsibly.
Sign up for free here now and meet the Mates. That way, you use artificial intelligence responsibly – and still stay in control of your communication at all times.
Conclusion
At first glance, the transparency obligations of the EU AI Act seem complex. But once you look at concrete use cases, a pattern quickly emerges: many classic marketing materials do not automatically need to be labeled. AI chatbots, deepfakes and certain publications on matters of public interest, on the other hand, deserve special attention.
For companies, it therefore pays to create clear processes for the use of artificial intelligence and to deliberately put people at the center.
This not only creates more legal certainty, but also more trust among customers, partners, employees and social media users.

