AI labeling starting in August 2026: What SMEs need to clarify now in their marketing efforts

It’s just two months away. On August 2, 2026, the labeling requirement for AI-generated content under the EU AI Act will take effect. Images, videos, audio files, and text from ChatGPT, Midjourney, Sora, ElevenLabs, and similar platforms must then be labeled as AI-generated. For marketing managers at small and medium-sized businesses, this isn’t some distant compliance task. It’s the next concrete deadline in their day-to-day operations.

We’ve been getting questions about this for weeks: “Do we have to label every ChatGPT image now?” “Is a footer note enough?” “What if we’ve only edited the text?” It’s high time for a clear assessment of the situation.

What exactly applies starting August 2, 2026

The requirement is set forth in Article 50 of EU Regulation 2024/1689, better known as the AI Act. The article establishes transparency obligations for providers and operators of AI systems. Implications for SMEs: Anyone who uses AI tools to generate marketing content is considered an operator under the regulation. Therefore, the requirement applies.

Four categories are affected (artificialintelligenceact.eu):

1. Chatbots on the website: Users must be able to recognize that they are interacting with an AI system.
2. Synthetic images, videos, and audio: Whether for social media, newsletters, or product pages.
3. Deepfakes: Realistic-looking AI images depicting people, places, or events.
4. AI-generated texts on public topics: blog articles, press releases, statements.

The requirement applies to companies of all sizes. There is no SME exemption. As soon as the content is published and is not clearly labeled as satire or art, the rule applies.

What many customers misunderstand

We hear three misconceptions particularly often:

First: "We only do this internally." Wrong. As soon as the content becomes visible externally, the requirement applies. An AI-generated product description in the store, an AI image in the newsletter, an AI text in a LinkedIn post: all are affected.

Second: “A footer note is enough.” Wrong. The label must be visible at the touchpoint. If you run a chatbot on your website, the notice needs to be in the chatbot window, not in the legal notice. An AI image needs a visible label on the image or directly below it, not in the page footer.

Third: “If we edit it, it’s exempt.” Half-true. Article 50 provides an exception for texts: the obligation does not apply in cases of editorial revision under human supervision. However, the image remains AI-generated. The audio remains AI-generated. The exception applies only where the human contribution significantly shapes the work. Someone who skims a ChatGPT text once and corrects a comma error does not fall under the exception.

Two Levels of Labeling

The AI Act requires two parallel labels (IHK Munich):

Machine-readable: In the file’s metadata. For images via EXIF/IPTC or the C2PA format (an industry standard for provenance). For audio and video, similarly via container metadata. Humans do not see this labeling directly, but search engines and platforms do.

Human-readable: Visible at the touchpoint. A "AI-generated" label below the image, a note stating "This voice was generated using AI" before the audio, or a disclaimer stating "This text was created using AI" in a blog article.

Anyone who meets only one of these criteria is not compliant. In May 2026, the Wettbewerbszentrale published a guide for companies (IHK Ostwestfalen) that emphasizes precisely this point.

What this means for everyday marketing

We have prepared a checklist for our clients. The list can be worked through in any SME in an hour:

The "How to implement" column is the real pain point. With WordPress, this can be solved using plugins or custom fields. With HubSpot, Mailchimp, or other marketing tools, this feature is often missing. Here, a team workflow is needed.

Three steps for SMEs by early August

We recommend a pragmatic approach. Three steps that can be accomplished without a marathon of consultations:

Step 1: AI Tool Inventory (1 hour).
Who uses which AI tools in marketing? ChatGPT, Claude, Gemini, Midjourney, DALL-E, Sora, ElevenLabs, Suno, RunwayML, Adobe Firefly. Create a list and document the use case for each tool. For each use case, ask: Is the output shared externally? If so, it must be labeled.

Step 2: Labeling Concept (2 hours).
Define what the labeling will look like for each output channel. Examples:
- Blog articles with AI content: A note box at the top or in the author section
- Newsletter images: Add "AI-generated" to the alt text and include a visible label -
Social media posts: Use the hashtag #AIgenerated or include a note in the caption
- Chatbots: Customize the welcome text

Step 3: Clarify responsibility (30 minutes).
Who is responsible internally? Marketing, IT, or management? This person must maintain the inventory and review new tools before they are introduced. In the event of fines, the company is liable, not the intern who wrote the prompt.

Fines: What’s Really at Stake

The AI Act provides for fines of up to 15 million euros or 3 percent of global annual turnover for violations of Article 50 (Bitkom Implementation Guide). Sounds like a penalty for large corporations. And it is.

More realistic for SMEs: warnings from competitors. The Competition Center didn’t publish its guidelines for no reason. Anyone who distributes AI-generated content without proper labeling risks receiving cease-and-desist letters from within their own industry. That costs fewer euros, but more stress.

The bigger picture

We do not view the labeling requirement in isolation. It is part of a wave set to hit companies in 2026:

• February 2025: Mandatory AI competency for all employees who use AI in their work (Article 4 of the AI Act). This has been in effect for over a year. Many SMEs have not yet documented this.
• August 2025: GPAI rules and penalty framework take effect.
• August 2026: Mandatory labeling of AI content (Article 50).
• December 2026: Technical labeling requirement with machine-readable watermarks.

Anyone who only addresses Article 50 now will have the next issue on their plate in four months. We recommend addressing both in parallel: inventory, proof of competence, labeling. One meeting per quarter is sufficient to keep things in order.

What we’re doing specifically

In our client projects, we’re currently adding two components to WordPress setups: a “AI content” custom field per post and an automatic labeling system for post images. Both are integrated cleanly into the theme without an external plugin. Effort per site: 3 to 5 hours for setup, then it runs on its own.

For social media, an internal briefing document usually suffices: Which tools are approved, how content is tagged, who reviews it. A single page of text that goes into the marketing onboarding checklist.

Conclusion

The labeling requirement starting in August 2026 is not a bureaucratic monster. It is a clearly defined step that can be implemented in any SME in half a day. Those who wait run the risk of receiving warnings from competitors shortly after the deadline. Those who act now have a six-week buffer.

If you use AI tools in marketing and are unsure where your obligations begin, have the potential of your own setups reviewed: Get in touch.