Since 2 August 2026 the transparency rules of the European AI Regulation, better known as the AI Act, apply. Anyone publishing AI images, AI video or AI text must make that clear in a number of cases. We get many questions about it, especially from business owners who generate images for their site or social media themselves. This is what the rules say, in plain language, and how we handle it ourselves.
What exactly has changed
The AI Act came into force in phases. Article 50 covers transparency and applies since 2 August 2026. It distinguishes between two parties:
- Providers of AI systems, the companies behind the models. They must mark their output in a machine-readable way, for example with a watermark or metadata. Systems already on the market before 2 August have a transition period until 2 December 2026.
- Deployers of those systems. That is you as soon as you generate an image and publish it. In certain cases you must visibly state that it is AI.
That second group is the relevant one for most businesses. The provider's marking does not release you from your own obligation.
When you have to label
The law requires a visible disclosure in three situations:
- Deepfakes. Image, audio or video generated or manipulated by AI that resembles existing people, objects, places or events, and that could appear real to a viewer. A generated photo of a real person falls under this. A photorealistic image of an office or a street can fall under it too, even if no real place was copied.
- AI text on matters of public interest. News-like texts written with AI and published to inform the public. If a human has reviewed the text and carries editorial responsibility, the obligation does not apply.
- Chatbots. A visitor must know they are talking to an AI system, unless that is obvious.
What is not required: a label on every image AI was involved in. An abstract mood image, an illustration or an icon nobody would take for a real photo does not need a disclosure. Content created and published before 2 August 2026 does not need to be labelled retroactively either.
What a label should look like
In June 2026 the European Commission published a code of practice, the Code of Practice on Transparency of AI-generated Content. It is voluntary, but whoever follows it may assume they comply with the law. The essentials for a website:
- The label must be immediately recognisable, without a visitor having to click or hover.
- It preferably sits in or directly next to the image itself, not only somewhere on a terms page.
- At minimum the word "AI" is needed; "AI generated" or "AI modified" is clearer. The Commission also offers free EU icons.
- An invisible watermark from the provider does not replace the visible label.
What it costs if you don't
Breaching the transparency obligation can lead to a fine of up to 15 million euros or 3 percent of worldwide annual turnover, whichever is higher. For small businesses proportionality is taken into account, but it is not a formality. More importantly: a client who discovers that a "photo" of your team or your workplace was generated will trust the rest of your site less too.
How we do it ourselves
On this website we use generated mood images for our services. They are not photos of our workplace, but images that set a tone. Legally it is a grey area: realistic, but of a place that does not exist. So we chose not to hesitate. Every generated image carries a small "AI image" label in the bottom right corner, and the footer states that project images are never generated by AI. Screenshots of delivered work are always real.
We apply the same rule for clients we use AI for. When we generate product images, blog illustrations or social posts, we build the label into the process, so nobody has to remember it afterwards. A generated text only goes online after someone has read and approved it, so editorial responsibility rests with a human.
Practical checklist
- List which images, videos and texts on your site and social media were made with AI since 2 August 2026.
- Judge each image: could someone take this for real? If so, label it.
- Put the label in or directly next to the image, small but readable, in every language of your site.
- Always have AI texts checked by a human before they go online and record that in your process.
- State that a chatbot is an AI.
- Document your choices, so you can show why something was or was not labelled.
A note
This article is not legal advice. The AI Act is new, the code of practice is voluntary and regulators have yet to rule on the first cases. If you are unsure about a specific case, ask a lawyer who knows the regulation. What you can do today is simple: be open about what AI has made. That is not only the law, it is also just decent towards your visitors.
