Labelling AI Images on Your Website: What Has Applied Since August 2026

Labelling has been mandatory since August 2026

The transparency obligations of Article 50 of the EU AI Act have applied since 2 August 2026. For a hotel or restaurant website, one point matters most: anyone publishing images that were generated or altered with AI and look like real photographs must make that visible. The obligation lies with the operator of the website – that is you, not the provider of the image generator and not automatically your agency.

The good news: not every AI image is covered. And with a clear process, implementation takes a manageable amount of time.

Which images you have to label

The regulation speaks of “deep fakes” and means more than faked celebrity videos. It refers to an image that meets three criteria at once:

• It was generated or substantially altered with AI.
• It shows people, places, objects or events that exist or could plausibly exist.
• A viewer would take it for a real photograph.

No intention to deceive is required. Typical cases in hospitality:

• the AI-generated “photo” of a room, the terrace or the breakfast buffet,
• a dish that was never photographed like this,
• people who do not exist, shown as guests or as your team,
• a real photo into which AI has added something – the pool that does not exist yet, or the mountain panorama behind the building.

Which images are not covered

• Recognisably artificial images: illustrations, graphics, abstract or deliberately stylised motifs that nobody would take for a photo.
• Standard image editing: exposure, cropping, noise reduction or retouching do not turn a photo into an AI image that must be labelled – even if the editing software uses AI features for it.

When in doubt, one simple question helps: could a guest expect something from the image that they will not find on site? Then you should label it – or better, use a real photo. Our quick check gives a first orientation in three questions.

What the label has to look like

The European Commission published a Code of Practice in June 2026 and guidelines on Article 50 in July 2026. The key points:

• The label must be clearly perceivable for people – at the moment they see the image for the first time.
• It therefore belongs on the image itself. A notice in the legal notice, in the terms of use or only in the file's metadata is not enough.
• The Commission provides standardised icons, including the “AI” symbol. Your own designs are possible if they are just as clearly recognisable.

What applies to older images

According to the Commission's guidelines, content generated before 2 August 2026 does not have to be labelled retroactively. Two reasons still speak for going through your existing images:

• It is often no longer possible to prove reliably when an image was created.
• The ban on misleading advertising under competition law has no cut-off date. An AI image that shows a room nicer than it is can lead to a warning letter regardless of the AI Act.

Five steps to a clean website

1. Review what you have. Go through the media library: which images come from a photographer, which from a stock library, which from an AI tool? Ask your agency as well – they know what they generated.
2. Classify. Decide for every AI image: generated, edited or not affected.
3. Label visibly. The label belongs on every occurrence of the image – including galleries, sliders and featured images.
4. Document. Record who classified which image and when. That helps when someone asks.
5. Define a process for new images. Who uploads images, who classifies them? Agree with your agency that AI images are delivered marked as such.

How to implement it in WordPress

Doing it by hand is tedious: lettering placed into the image file has to be redone for every crop, and a caption is missing wherever the theme does not output it.

That is why we developed the plugin Ascensus AI Labelling. You classify an image once in the media library – “AI-generated” or “AI-edited”, also many images at once. The label then appears automatically wherever the image is used: in Elementor, in the block editor and on featured images. Every classification is logged.

There are two things the plugin deliberately does not do: it does not detect AI images by itself, and it does not make the decision for you whether an image has to be labelled. The sample events on our EventBoard page show what a labelled image gallery looks like: all motifs there are AI-generated and marked accordingly.

What happens in case of violations

For violations of the transparency obligations, the AI Act provides for fines of up to 15 million euros or 3% of worldwide annual turnover. That is the upper limit, not the normal case for a family business. In Germany, the supervisory authority is the Federal Network Agency (Bundesnetzagentur).

In everyday life another risk is closer: every competitor can see a missing label with the naked eye, and a warning letter for misleading advertising is quickly written.

This article provides orientation and does not replace legal advice. The relevant provisions are Article 50(4) and Article 3(60) of Regulation (EU) 2024/1689 and the European Commission's guidelines. In individual cases, ask your legal adviser.