AI Governance
Article 50 AI Act: What Companies Publishing Articles and Posts Need to Do
Article 50 of the EU AI Act introduces new transparency duties for certain AI-generated and AI-manipulated content. This article explains what companies publishing articles, posts and other public-facing materials should label, how disclosure should be presented, and what practical steps organisations should take to remain compliant.
On 2 August 2026, the transparency requirements under Article 50 of the AI Act came into force across Europe. The Commission has released the final Guidelines and Code of Practice, along with a set of official icons for labelling artificially generated content. For most companies that regularly publish articles and posts, this is not about “new fines” — it is a practical question of when you need to clearly state that a text was created with the help of AI, and when you do not.
Let’s break it down.
What exactly is required from those who publish materials
The labelling obligation primarily concerns deployers — those who use generative systems and make the output publicly available. Two main cases fall under the rules:
deepfakes (images, audio or video that appear authentic);
texts on matters of public interest that have been generated or substantially modified by AI and have not undergone proper human editorial review.
Analytical articles and expert posts almost always qualify as “matters of public interest.” This means that if a piece appears on a website or social media that was written almost entirely by a neural network, and a person only briefly glanced over it, labelling is required. If the author has properly revised the text, checked the facts, added their own conclusions, and the company assumes responsibility for the content — labelling is not required. This is the most important exception currently used by most careful companies.
How to use the official icons
The Commission has published a free set of icons (SVG and PNG, several colour variants). There are three main types:
Fully AI-Generated — when the text was created entirely by artificial intelligence;
Partially AI-Modified — when a person wrote the base and AI substantially reworked it;
Basic icon — a universal option, especially useful when accompanied by a text explanation.
Combinations of an icon with short text work best. A plain pictogram without explanation is less effective.
In practice it looks straightforward. At the beginning of an article or post (right after the title or even before it) a small block is placed:
“This article was generated with the help of artificial intelligence”
or
“This material was partially created using AI”
The icon should be visible immediately, without the need to scroll. It is also preferable that the label remains when the content is reshared or downloaded.
Official documents and icons can be found here:
Guidelines on Article 50
Code of Practice and icons
Ready-to-use wording
Fully generated text:
“This article was generated with the help of artificial intelligence. The company has not conducted full editorial review and does not assume responsibility for the content.”
Partially modified:
“This article was prepared with the assistance of artificial intelligence. The text was partially generated or substantially modified. Final review was / was not performed by the author.”
Short versions for the header:
“AI-generated”
“Partially modified with AI”
What a company should do right now
The most straightforward approach is to introduce an internal rule: any publication issued in the company’s name undergoes proper editorial review, and the company assumes responsibility for the text. In that case, labelling will rarely be needed.
If you occasionally publish “raw” AI content (for example, quick overviews or posts), build a ready-made block with the icon and disclaimer into your website templates and post templates. This is cheaper and simpler than having to explain yourself later.
Signing the Code of Practice is not mandatory for small and medium-sized companies. It offers a certain convenience during checks, but at the same time requires strict adherence to all its technical details. Many companies are currently simply complying with the legal requirements without formally joining.
Conclusion
Article 50 does not prohibit the use of AI for preparing articles and posts. It requires transparency in those cases where a person has effectively stepped back from responsibility for the text. For any company that publishes content, this is a good opportunity to put editorial processes in order: either edit properly, or label honestly.
The official icons and Guidelines are already available. It remains only to decide how exactly you will work with them on your website and social media.
This material is for informational purposes only and does not constitute legal advice.
