New EU Rules on Labelling Synthetic Content: What It Means for Your Website

What happened and why it matters

The European Union has adopted new rules on the transparency of AI-generated content. From August 2, 2026, labelling AI-generated content will be a legal requirement for every business that serves EU citizens. Similar to the European Accessibility Act, these rules don't apply only to European companies, but to any company in the world whose content is used by people in the EU.

Even though panic and rumors about "huge fines" and "drastic measures" have been spreading online this week, the truth is much narrower and more reasonable. The core of the rules is simple: when AI-generated content could easily be mistaken for something made by a human, the creator must clearly and unambiguously say so. It's not about labelling every single word a tool helped you write, but about not misleading the user.

In this article, we explain exactly what needs to be labelled, where the line falls between "edited" and "generated" content, and what this concretely means for a Slovenian small or medium-sized business and its website.

What needs to be labelled

The goal of labelling is for anyone who encounters AI-generated content to clearly and recognizably identify that the content was generated or manipulated by artificial intelligence. Under Article 50(4) of the AI Act, labelling applies to:

• Deepfakes. Any image, audio, or video that resembles a real person, object, place, or event and would falsely appear to be authentic or real. Content that isn't convincingly realistic generally doesn't fall under this category.
• Chatbots and AI agents. Users must be informed if they are not talking to a human.
• Fully AI-generated text. Especially on matters of public interest, where there has been no human review or editorial work.
• Emotion recognition and biometric categorization tools.

Important: legal obligations apply to both providers (who build or supply the tool) and deployers (who use it). Similar to GDPR, a company cannot avoid Article 50 simply because it licensed the tool from a third party. You, not the tool provider, are responsible for the content on your website.

What does NOT need to be labelled

This is the most important part for most businesses: the vast majority of content where you simply got some help from a tool doesn't fall under the new transparency rules at all.

The disclosure obligation does not apply when AI-generated text has been reviewed and edited by a human, with a named person or company taking editorial responsibility for it. In other words, if someone actually reads your text, corrects it, and takes responsibility for it, you're in the clear.

Confusion tends to arise around the meaning of "public interest." This refers to health, safety, the environment, the economy, finance, politics, science, and culture. If AI-generated claims about a product touch on these areas, disclosure applies. Some law firms recommend labelling realistic AI-generated illustrations or photos in advertising and marketing just to be safe. AI-generated product photos or posters require disclosure if they resemble a real person, place, object, or event.

The thin line between "edited" and "generated"

The key question for a business owner is: where does editing end and generating begin? The European Commission's guidelines are somewhat vague on this point, but some things are clear.

Minor supporting edits do not count as generation:

• spelling and grammar checks,
• formatting,
• cropping and color correction of images,
• machine translation.

The following do count as generation:

• AI-generated summaries,
• composite (combined) images,
• content-level rewrites of an entire text,
• adding or removing elements from a photo.

In practice, this means: fine-tuning a sentence a human wrote is fine, but generating an entire sentence from scratch requires disclosure. The Commission is explicit that "someone quickly skimmed the text before publishing" does not count as editorial review. The review must be substantive, with a named responsible person. So the line runs between deliberate manual intervention and automatic generation, with the latter always requiring disclosure (except in closed B2B business environments).

A sparkle icon probably isn't enough

Many tools today use a generic sparkle icon to mark "AI-powered features." As part of its Code of Practice, the European Commission has published a specific set of EU icons for labelling AI-generated content, which is not the same as this sparkle symbol.

The signal must be "clear and distinguishable." A sparkle icon is too ambiguous, since it often means "this feature is powered by artificial intelligence" rather than "this specific piece of content was generated by a tool." The Commission is explicit that using an icon does not by itself mean legal compliance. The following are considered non-compliant:

• a barely visible icon,
• a note hidden in the page footer,
• a label that appears for only a second.

The icon should be clearly visible, accompanied by a text label in understandable language ("AI-generated"), and accessible to assistive technologies. The label must also persist when content is shared or downloaded further. A recommended safe choice is pairing the icon with clear text.

This isn't just an EU matter

You might think this is another one-off European measure, but it's actually part of a global pattern:

• China has had mandatory labelling since September 1, 2025, with visible labels and watermarks in metadata.
• California has SB 942 (amended by AB 853), which takes effect on the same day as the EU rules, August 2, 2026, deliberately aligned.
• South Korea has a foundational AI law, in effect since January 22, 2026.
• India has amended its IT rules, requiring labelling of "synthetically generated information" from February 20, 2026.

When nearly the same rule appears on four continents within a year, that's not a coincidence. If you plan to publish anything with the help of these tools this year, it's worth thinking now about what will need labelling and what won't.

What this means for your business and website

For most Slovenian small and medium-sized businesses, there's no reason to panic, but there is reason to get organized. In practical terms, this means:

1. Make an inventory. Where on your website and in your marketing do you use AI-generated product photos, realistic illustrations, or text without human review? These are exactly the risky spots.
2. Establish editorial responsibility. If your blog posts, product descriptions, and news are actually read and edited by a named person who takes responsibility for them, you're mostly not required to disclose. This is a cheap and effective solution that also has other benefits for content quality.
3. Review your chatbots. If you have a chatbot on your site, visitors must clearly know they're not talking to a human.
4. Align your labels. If you're already labelling content, make sure the label is clear, visible, readable by screen readers, and persists when shared.

This isn't a developer problem, it's a business risk and reputation issue. A wrongly placed or hidden label can mean non-compliance, while transparency builds customer trust in the long run. People are increasingly distinguishing between quality human-made content and a flood of generic text, and transparency helps you stay on the right side of that line.

If you're not sure what on your site falls under the new rules and how to handle it without technical confusion, at Carpolab we can review your website, content, and labelling approach and set up a compliant, accessible solution. Get in touch and we'll map out the risky spots together.

Frequently asked questions

Do I need to label every article a tool helps me write?

No. If the text is reviewed and substantively edited by a human who takes editorial responsibility for it, disclosure is generally not required. However, a quick skim before publishing doesn't count as genuine editorial review.

Do the rules apply to me even if I'm not based in the EU?

Yes, if your content is used by people in the EU. The rules apply to any business in the world whose output is used by EU citizens, similar to the Accessibility Act.

Is it enough to just add a sparkle icon?

Probably not. A sparkle icon is too ambiguous, and using an icon alone does not guarantee compliance. It's safer to pair the icon with clear text such as "AI-generated" that is visible, accessible, and persists when shared.

---

Building a new website or redesigning an existing one is a great opportunity to build in transparency and accessibility from the start and avoid fixes later on. If you're curious how much this would cost for your situation, check out our calculator for a rough estimate.