Since August 2, 2026, a new part of the EU AI Act, the European AI Regulation, is in effect throughout the EU. It regulates when people must be informed that they are dealing with artificial intelligence. This does not only concern tech companies. Any business with a chatbot, AI-generated images, or AI-written texts on their website should take a quick look.
The good news upfront: For most small businesses, the effort is manageable. In this article, I explain what applies, what does not apply, and how you can check in fifteen minutes whether you need to take action.
Important: I am a web developer, not a lawyer. This article is a practical overview, not legal advice. In case of doubt, please consult a specialized lawyer.
What changed in August
The EU AI Act is being implemented gradually. On August 2, 2026, the so-called transparency obligations (Article 50) became mandatory. Shortly before that, on July 20, the EU Commission published guidelines that explain more precisely how these obligations should be interpreted.
In case of violations, fines of up to 15 million euros or 3% of global annual revenue are possible. For small and medium-sized enterprises, the lower of the two amounts applies. Therefore, no one needs to fear losing their business due to a missing notice in the chat window. However, you should not ignore the topic.
The most important question first: provider or operator?
The law distinguishes between two roles:
- Provider are those who develop and market an AI system, such as manufacturers of chatbot software or image generators.
- Operator are those who use such a system for their own purposes. If you integrate a ready-made chatbot on your website or create images with an AI tool, you are in this role.
This distinction is crucial because the responsibilities are divided. A large part of the technical work lies with the provider. For you as an operator, visible notices at the right places remain important.
Four typical cases on your website
1. The chatbot in customer service
Anyone writing with an AI chatbot on your website must be able to recognize that they are not speaking with a human. Formally, this obligation lies with the provider of the chatbot. In practice, you should still check whether the notice on your page is truly visible.
A special case is a chatbot that is not a finished product but has been developed individually for you, for example, via the interface of an AI model. In this case, you can be considered a provider yourself, and the obligation to provide notice lies directly with you. Discuss this with your developer.
There is an exception when it is "obvious" that it is AI. However, this exception is narrowly defined. A support bot that is fluid and human-like is not considered obvious. If visitors could believe they are writing with a person, you need a notice.
The solution is simple: a short sentence at the beginning of the chat, such as "You are chatting with our AI assistant." Ideally with the option to reach a human.
2. AI-generated images and videos
There are two levels here:
The technical marking is handled by the tool provider. AI images must be marked in a machine-readable format, for example through metadata or invisible watermarks. You don't need to worry about that.
The visible labeling concerns you, but only in the case of so-called deepfakes. This refers to images, videos, or audio recordings that realistically replicate real people, places, or events to the extent that they could be mistaken for real.
An AI-generated stock image of a coffee cup is not a deepfake. A realistic photo showing your team at an appointment that never took place would be one, however. Easier rules apply for clearly artistic or satirical content.
My advice: If in doubt, label voluntarily. A small note "Image: AI-generated" costs nothing and builds trust. The EU even provides standardized symbols for this.
3. AI-written texts
Here, the obligation is much stricter than many think. You only need to label AI texts if they are intended to inform the public about matters of public interest. The typical company blog about your services or your industry is unlikely to fall under this.
And even then, the obligation is waived if a human has reviewed the text and someone takes editorial responsibility. This is how AI should be used in writing anyway.
By the way: Spell checking, linguistic smoothing, and translations with AI are considered normal editing. AI summaries and content rephrasing are technically marked by the tool provider. This does not change anything for you.
4. Content that is already online
Content created before August 2, 2026, does not need to be retroactively labeled. There is also a transition period for technical marking until December 2, 2026. However, this only applies to AI tools that were already on the market before August 2, 2026. Newer tools must mark their content from the start.
So you do not have to scour your entire website. It is about what you publish from now on.
What the law does not require
There is a lot of uncertainty circulating, so to be clear:
- The AI Act does not prohibit you from using AI for your website. Article 50 requires transparency, not abstinence.
- You do not have to label every text and every image with a warning.
- The strict rules for so-called high-risk AI affect areas such as personnel selection, credit granting, or critical infrastructure, not the website of a craft business or a practice. They have also been postponed to December 2027.
A requirement has been in place since February 2025 and is often overlooked: Businesses that use AI should support their employees in competent handling of it (Article 4). Since a change in the law in July 2026, this requirement has been softened. For small businesses, this mainly means in practice: The team should know what AI is used for and where its limits lie.
Checklist: Checked in 15 minutes
- Chatbot: Is it clear at the beginning of the conversation that it is AI? Was it developed individually? Then clarify who is considered the provider.
- Images: Are there AI-generated images that deceptively represent real people, places, or events? Then label them.
- Texts: Do you publish AI-generated texts on topics of public interest without human review? Then either review or label them.
- Tools: Do you use AI tools from reputable providers that implement technical labeling?
- Responsibility: Is it clear who in your company approves content before it goes online?
- Team: Do your employees know how and for what AI is used in the company?
- Data Protection: Does your chatbot process personal data? Then it belongs in your privacy policy. This is a different law, but it is often forgotten.
How I work with AI myself
I use AI daily in my work on websites. It significantly speeds up many steps. This ultimately makes a website cheaper for my clients, and with the time gained, I can invest more in quality: in details, in testing, in a clean structure.
The decisive factor is: The responsibility remains with me. AI provides drafts, I review, correct, and decide. I have described why this is particularly important with code in a separate post.
And because it fits the topic: This post was also created with AI support. A research assistant took care of the research and rough draft for me. I have reviewed, adjusted the content, and take responsibility for it myself. Legally, this note is not necessary here. However, I believe it is part of an honest way of working.
Conclusion
The new labeling requirement is not a cause for concern, but a good opportunity to tidy up briefly. A note in the chat window, a remark under realistic AI images, and a clear approval for texts: That's all most websites need.
Transparency is not only a requirement but also an advantage. Those who deal openly with AI appear more credible than someone who hides it.
Read more:
- Website from AI code: Why a generated draft is not a foundation
- What makes a good website today – completely without jargon
Do you want to know if your website meets the new requirements, or are you planning an AI chatbot? I would be happy to take a look at it with you. Through “Inquire about the project” you can reach me without obligation.