On August 2, 2026, the last major component of the AI Act will take effect. As soon as you produce images, texts, or chatbots with artificial intelligence, this will affect you just as much as the large tech corporations. In this article, we clarify which of your content will require a notice from now on and which will not.
What the AI labeling requirement from August 2, 2026, means for your marketing
From August 2, 2026, the AI labeling requirement will apply throughout the European Union. Anyone who generates images, operates a chatbot, or produces texts with language models must disclose from that day on what originates from a machine. Because many guides generally speak of millions in fines, there is great uncertainty among small and medium-sized businesses. The actually important question remains unanswered: Which content is actually affected?
At VASTCOB, we support companies through precisely such regulatory upheavals, currently primarily in AI-supported marketing processes. To save you from having to piece together the situation yourself, this article categorizes the rules, defines the limits of the obligation, and provides you with a checklist for implementation.
The AI labeling requirement starts on August 2, 2026
Behind the AI labeling requirement is Article 50 of the European AI Regulation. Anyone who offers or professionally uses AI systems must make artificially generated or altered content recognizable. The enumeration shows how far this extends: chatbots, synthetic images, audio and video files, and certain text forms.
The basic idea is simple: no one should unknowingly speak to a machine or mistake a fabricated recording for real.
The AI Act roadmap until this deadline
The AI Regulation has been in force since August 2024, and it will be activated in stages. First came the prohibited practices, then the rules for general AI models. When the AI labeling requirement in 2026 is mentioned, it refers to the third stage, the transparency obligations. Anyone working in marketing or communication will feel this stage first.
Why the postponed deadlines don't change anything here
In the course of the so-called Digital Omnibus, the EU has postponed several deadlines, including the requirements for high-risk systems. Anyone who concludes that the AI Act has been completely postponed is mistaken. This is not true for the transparency rules.
What is summarized in practice under the AI Act labeling requirement remains at the original date. Only one detailed rule is moved back, namely the machine-readable marking by providers of systems that were already on the market before. If a system was already on the market before, this obligation is postponed by four months.
Who must label according to the AI Act and who does not
The regulation distinguishes between two roles. Almost the entire AI labeling requirement depends on this distinction. Providers develop AI systems and bring them to market, operators use these systems professionally. As soon as you work with image generators, language models, or chatbot software, the second role is usually relevant for your company.
Providers and operators in distinction
Providers must ensure that generated content is technically recognizable as synthetic, for example, through watermarks or metadata. Since this level runs in the background, it is the responsibility of OpenAI, Google, Adobe, and co.
Become more visible on Google & Social Media?
In a free strategy consultation for data-driven online marketing, we uncover your untapped potential, review any existing ad accounts if necessary, examine your SEO ranking and visibility, and determine which strategy is appropriate for your budget and which active measures will lead to more inquiries or sales.
✅ More visibility & perception through targeted placement
✅ More visitors > prospects > customers > revenue
✅ Reach target groups scalably with SEA
✅ Act and grow sustainably with SEO
🫵 Maximum success with our hybrid strategy
💪 More than 15 years of experience across industries in over 1,000+ projects demonstrable!
Your duty as an operator begins where you publish the finished content. Those who clearly separate the roles save themselves a lot of unnecessary worry, because only upon publication do you have to label AI-generated content.
Why even small businesses are obliged
The AI labeling requirement does not have a revenue threshold. As soon as content is published professionally and made publicly accessible, a craft business with twelve employees falls under it just as much as a corporation. However, the size of the company does play a role in the amount of possible sanctions. Purely private use remains excluded, as do internal drafts that never leave the premises.
If you are unsure which systems in your company fall under the regulation at all, we will clarify this in an AI consultation with a structured inventory.
What content the AI labeling requirement covers
Anyone who enters the search term „labeling requirement AI-generated content“ expects a blanket rule for everything that comes from a model. Article 50 does not go that far. The regulation names four clearly defined categories of cases. Outside of these, there is no legal obligation to provide a notice.
Chatbots and AI assistants in customer contact
Anyone who uses an assistant on their website must let the user know that there is no human answering at the other end. A sentence in the greeting is sufficient for this.
The exception applies when the situation is clear anyway, for example, with a widget that is visibly named as an AI assistant. Thus, the AI labeling requirement affects precisely the channel through which many companies first work with artificial intelligence.
Synthetic image, audio, and video content
For synthetic media, the regulation separates two levels. While the provider ensures machine-readable marking in the file format, the operator takes over the visible notice. This primarily refers to deepfakes, i.e., deceptively real representations of actual persons, places, or events. Anyone who wants to label AI-generated content should logically start with these realistic formats.
AI texts on topics of public interest
For texts, the AI labeling requirement is much narrower than many assume. If a contribution informs the public about matters of public interest, it is covered.
An editorial review with human responsibility removes the obligation again. Since product descriptions, newsletters, and service texts do not fall under it anyway, an SEO text supported by AI from a clean editorial process also remains unproblematic.
| Content type | Mandatory | Responsible | Legal basis |
|---|---|---|---|
| Chatbot or voice assistant | Notice of AI interaction | Provider | Art. 50(1) |
| Synthetic images, audio, video, text | machine-readable marking | Provider | Art. 50(2) |
| Emotion recognition and biometric categorisation | Information of the data subject | Operator | Art. 50(3) |
| Deepfake | visible disclosure | Operator | Art. 50(4) |
| AI text on matters of public interest | visible disclosure | Operator | Art. 50(4) |
Do AI-generated images need to be labeled?
Do AI-generated images need to be labeled as soon as they appear on the website or in the newsletter? No, at least not across the board. How realistic the subject appears determines whether a notice is required, not the tool behind it.
When an image counts as a deepfake
Under the heading of AI Act labeling requirement for images, very different interpretations are currently circulating. What is meant are depictions that look deceptively similar to a real person, a real place, or an actual event. As soon as a viewer could mistake the subject for a real recording, you need a notice. With a stylized illustration, this threshold is not reached.
These images require a notice:
- Photorealistic people who do not exist, for example as a testimonial or team photo
- Re-enacted real situations with a documentary feel
- Altered real footage where content has been added or removed
- Voices and video sequences of real people
These images generally do not require one:
- Clearly illustrative graphics, icons, and abstract patterns
- AI-powered retouching without altering the content of the motif
- Internal drafts and mood boards without publication
- Artistic or satirical works, for which a special rule applies to minimize impact
Which visuals do not require a notice
We have seen how much time gray areas cost in daily work in over 1,000 implemented projects. Therefore, we prefer to interpret the AI labeling requirement generously in case of doubt. While a subsequent discussion about deception costs trust, a brief note in the image caption practically never diminishes the advertising impact.
How to label correctly without losing your impact
The information required by the AI labeling obligation must be clear, recognizable, and accessible. It must be present as soon as someone perceives the content for the first time or interacts with the system. A note in the imprint does not meet this requirement. The notice belongs where the content is consumed.
Formulations that are understandable and unobtrusive
When labeling AI-generated content, short and factual sentences without an apologetic tone have proven effective:
- Image caption: This image was created with AI.
- Video: An overlay in the first few seconds, and additionally in the description for sensitive topics.
- Chatbot: A greeting sentence that names the AI support.
- Text: A note at the beginning or end, provided the case group applies at all.
Machine-readable markings such as watermarks and C2PA
Parallel to the visible notice, the technical level of the AI labeling requirement is also in play. Content Credentials based on the C2PA standard store origin data directly in the file, and many generators now set them automatically. If your workflow discards this metadata during export or when uploading to the content management system, it is lost right there.
The right time for the notice
Since transparency and brand impact are not mutually exclusive, a notice costs you nothing. Those who communicate openly appear confident; those who hide the notice create the impression of a concession. Treat the AI labeling requirement like a source citation, as a natural part of clean communication.
AI labeling requirement on Instagram and other platforms
In practice, you usually encounter this topic in a platform-specific way: AI labeling requirement on Instagram, TikTok, or YouTube, depending on where your content runs. How the networks implement this differs, sometimes automatically, sometimes as a self-declaration.
What Meta requires with the AI info label
Instagram and Facebook detect AI content via C2PA metadata and then automatically apply the AI info label. For photorealistic video and audio content, Meta additionally expects a self-declaration. Stricter rules apply to ads, which particularly affects companies with ongoing campaigns. This means Meta covers part of what the AI labeling requirement already demands.
| Platform | Labeling method | Automatic detection | Special feature |
|---|---|---|---|
| Instagram and Facebook | „AI info“ label plus self-declaration | C2PA metadata | stricter rules for ads |
| TikTok | Toggle for AI-generated content during upload | C2PA and AI effects | certain AI formats generally prohibited |
| YouTube | Disclosure in the upload dialog | partially | Prominent label for sensitive topics |
| Content Credentials | C2PA | Text notice expected in business context |
TikTok, YouTube and LinkedIn Compared
TikTok works with its own toggle in the upload process and prohibits certain AI formats regardless of any labeling. When uploading to YouTube, you provide the information in the upload dialog, and the label appears particularly prominently for news, elections or health topics.
On LinkedIn, Content Credentials are read out, and in the business context a linguistic notice is additionally expected. Anyone managing multiple channels should therefore handle the AI labeling requirement centrally rather than per platform.
Why the platform label does not replace the legal obligation
Platform rules are house rules, not law. Simply flipping the switch does not automatically fulfill the AI Act labeling requirement; conversely, violations of the house rules risk loss of reach or suspension.
Since 2010 we have accompanied channels across platform changes and, as a Meta Business Partner, we see how quickly such requirements change. How we implement this in ongoing support in Social Media Marketing—a uniform standard across all channels saves the most coordination.
What threatens in case of violations and how you can protect yourself
The circulating figure of 35 million euros causes the most excitement and the most misunderstandings surrounding the AI labeling requirement. Since this framework applies exclusively to prohibited practices under Article 5, i.e. to things like social scoring, it does not apply at all to missing notices under an image.
Fines placed in realistic perspective
For violations of the AI labeling requirement, the regulation provides for up to 15 million euros or 3 percent of worldwide annual turnover. Because sanctions must also be proportionate, this noticeably relieves smaller companies.
Realistically, a first omission is more likely to result in a request for remediation than an existence-threatening fine. In the case of misleading advertising, competition law risks are added, which you should also take into account.
The Bundesnetzagentur as supervisory authority in Germany
With the AI Implementation Act of June 2026, the Bundesnetzagentur becomes the central market surveillance authority. Since it also serves as a complaints office, the competence center based there also accepts tips from the public. Complaints from competitors or customers are therefore likely to become the most common trigger for a review of the AI labeling requirement. A robust supervisory practice will only emerge in the coming months.
Your Checklist for the Coming Weeks
- Inventory of all AI-supported tools and content in the company
- Assign responsibility, ideally to a person in marketing or legal
- Define standard formulations for image, video, chat, and text
- Integrate a review step into the existing approval process
- Document exceptions so that decisions remain traceable later on
If you do not want to set up this process on your own, we as an AI agency support you in implementing the AI labeling requirement. Customers describe how this works in collaboration in their reviews on ProvenExpert. If you would like a non-binding assessment, you can reach us at any time via the contact form.
Frequently asked questions about the AI labeling requirement
Does the labeling requirement also apply to internal content?
No. The AI labeling requirement applies to professional publication vis-à-vis third parties. Presentations for an internal meeting, drafts in the team chat, or test images during the concept phase remain unaffected, as long as they do not leave the company.
Is a notice in the legal notice sufficient?
No. Where the content is perceived, the AI labeling requirement demands the notice, at the latest at the moment of first perception. Because a collective notice on a subpage is neither clear nor recognizable, it is not sufficient.
What happens to content published before August 2, 2026?
For already published posts, the AI labeling requirement does not require retroactive labeling. For providers of AI systems that were on the market before the cutoff date, however, a transitional rule applies. They have until the beginning of December 2026.
Do I have to disclose AI support when writing blog posts?
In most cases, no. As soon as a human reviews the text for content and assumes editorial responsibility, the AI labeling requirement no longer applies. For largely automatically generated posts on topics of public interest, the situation is different.
What will follow in the coming months
The cutoff date marks a beginning, not an end. On December 2, 2026, the machine-readable marking will follow for older systems, and at the same time it will become clear how strictly the Federal Network Agency interprets the new AI labeling requirement.
Anyone who defines a standard for their company now will not have to make adjustments later. Because audiences and business partners increasingly expect openness, the disclosure of AI use will foreseeably shift from a mandatory requirement to a quality feature.
DE
