This article helps assessing whether you need to label the content you publish via Swat.io as AI-generated. It covers only content published through Swat.io: All post types in the Publisher as well as your community management replies in the Swat.io Inbox.
When does this apply and to whom?
The transparency obligations under Article 50 of the EU AI Act apply as of the August 2nd 2026. They govern when and how AI-generated or AI-modified content must be disclosed. In a small number of cases as the party publishing the content (called the "deployer") you will need to make clear that content was created with the help of AI.
The binding law comes directly from the European Commission. The versions adopted in final form on 20 July 2026 are authoritative:
The page Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems explains the scope and application of the obligations;
The Code of Practice on Transparency of AI-Generated Content describes the technical labelling.
The short summary
For the vast majority of what you publish via Swat.io, you will not need a label. If you use AI to draft or refine a text and a human reviews that text before posting, no labelling is required. That covers everyday work: Post-captions in Swat.io Publisher and replies to comments in Swat.io Inbox. Only two cases actually trigger the labelling obligation:
A photorealistic AI image, video or audio that could pass as real (see Case 1);
AI-generated text on a matter of public interest, published without human review (see Case 2).
What about chatbots?
Article 50 also requires disclosure when an AI speaks directly with a person in a genuine dialogue. This does not apply to Swat.io. The AI suggestions in the Inbox are reply suggestions that a human always reviews and sends himself. So no one interacts directly with AI and no disclosure obligation arises. In case of automations, which can be configured, so that incoming DMs can trigger automatic replies, are not a chatbot in this sense either. The replies are drafted by humans beforehand and are in that case only activated automatically.
Case 1: Photorealistic AI media (deepfakes)
You need a label as soon as an AI image, video or audio looks so real that it could be mistakenly taken as authentic. The law calls this a "deepfake". Three features must come together:
Resemblance: The content closely resembles a real person, object, place or event;
Existence: The subject exists in reality or could plausibly exist;
Potential to deceive: The content could create the impression of being real or true.
A typical example is a photorealistic AI portrait of a person who does not exist, in a post. On the other hand clearly stylised, drawn or cartoon-like AI images do not fall under this, no matter how heavily AI was involved in the process. Context also matters: If your audience does not expect the depiction to be real anyway, the potential to deceive may not apply.
Note: You must give the notice clearly and perceptibly at the latest on first exposure, for example as a visible label in the image or an audible notice in the audio. It is not enough to rely on the invisible machine-readable marking from the AI provider. That technical marking is the provider's responsibility, the visible disclosure is yours.
You can find more details in Article 50: Transparency Obligations for Providers and Deployers of Certain AI Systems and in the Questions and Answers from the European Commission.
Case 2: AI text on a matter of public interest
This case is mainly relevant if you work in the public sector or publish content for an NGO. AI-generated or AI-modified text only requires labelling if all of the three following conditions are met. The text is:
Published;
Intended to inform the public;
Related to a matter of public interest.
Such matters include, among others, politics and democratic processes, public administration and services, justice, fundamental rights, public safety, public health as well as environmental protection.
The key exemption: Text that has undergone genuine human review or editorial control does not need to be labelled. What is meant is a substantive review by a person with the relevant expertise who takes responsibility for the content. A purely formal check such as spelling or grammar correction is not sufficient. In practice this means, that as soon as someone on your side approves the text on the merits and takes responsibility for it, the exemption applies.
Ordinary marketing, sales or support texts do not fall under this case anyway, because they do not inform the public about a matter of public interest.
How to label correctly
If a label is required, keep it factual and concise. Established, low-key wordings are:
"AI-generated": The neutral standard for fully AI-created content;
"AI-assisted": For content the AI helped with;
"Drafted with AI" or "Entwurf mit KI": Signals that a human reviewed and edited the content.
For short UI texts such as labels or tooltips, "AI-generated" or "AI-assisted" feel the most neutral, because they sound factual rather than like a warning.
The European Commission also provides a free, official icon set (an "AI GENERATED" bar, an "AI MODIFIED" bar as well as a simple round "AI" symbol). The bare "AI" symbol is the legal minimum, the combination of icon and short text is recommended.
Placement:
For image and video, the label belongs directly in the content, visible on first contact and long enough to be noticed;
For text, the notice needs to be near the headline;
For audio only, an audible notice is needed;
A note in the imprint or the small print alone is never sufficient.
Deadline and transition period
The transparency obligations apply from 2 August 2026. Two things you should know:
No retroactivity: Content published before 2 August 2026 does not need to be labelled retroactively. If such content is republished or substantially reworked after that date, it does fall within the scope though;
Transition period for the technical marking: For AI systems that were already on the market before 2 August 2026, the machine-readable marking obligation (Article 50, § 2) applies from 2 December 2026.
You can find the full legal text of the AI Act (Regulation (EU) 2024/1689) on EUR-Lex.
Important legal notice:
This article is provided solely for your information and is to be seen as general guidance, not a legal advice. Swat.io has no responsibility or liability for the legal assessment of your specific content. Whether a labelling obligation applies in an individual case depends on your specific situation. If in doubt, clarify legal questions with qualified legal counsel.