Since 2 August 2026, the transparency obligations in Article 50 of the EU AI Act apply. Since then, many businesses have received emails with subject lines like "Fines of up to €15 million – is your website affected?". The truth lies somewhere between panic and a shrug. This article sorts out what really matters for a typical small-business website.
One thing first: this is a practical overview, not legal advice. For edge cases, please ask a lawyer.
What has applied since 2 August
Article 50 covers four things. First: anyone talking to an AI system, such as a chatbot, must be able to tell, unless it is obvious anyway. Second: providers of AI generators must technically mark generated images, video, audio and text, for example through metadata. Third: anyone who publishes deepfakes must disclose that they are artificially generated. Fourth: AI-generated text meant to inform the public on matters of public interest must be labelled, unless a human has reviewed it and holds editorial responsibility.
The "Digital Omnibus" package adopted in July 2026 mainly postponed deadlines for high-risk AI. It did not change these transparency duties. The one relief: generators already on the market before August have until 2 December 2026 for the technical marking. That concerns the makers of the tools, not your website.
Chatbot on your website: yes, label it
If an AI assistant on your website answers questions or takes bookings, it must be clear that a machine is replying. One sentence at the start is enough: "I'm the AI assistant of Salon X. For a personal answer, just message us directly." It costs nothing and builds trust along the way.
Good chatbot providers now build this in by default. Still, check what your bot says in its first sentence – towards your customers, you are the one responsible.
AI images: this is where it gets tricky
An abstract background or an obvious illustration from an image generator does not trigger a labelling duty for you as the website operator. Realistic images that show real places, people or things and look genuine are different: the "photo" of your restaurant, your "team", a dish that never looked like that on a plate.
The AI Act defines a deepfake as artificially generated content that resembles existing persons, places or objects and would falsely appear authentic. An AI-generated "photo" of your own terrace can fall right under that. Then there is competition law: advertising with images that show reality more beautiful than it is risks a cease-and-desist letter for misleading advertising.
Our advice: for everything customers will later see in real life – rooms, team, products, food – use real photos. Use AI images only for mood and illustration, and when in doubt add a short "AI-generated" note in the caption.
AI text: mostly relaxed
Service descriptions, about pages and blog posts written with AI help, but reviewed and owned by you, generally don't need a notice. The text labelling duty targets content meant to inform the public on matters of public interest and published without human control. Marketing copy for your hair salon hardly falls under that.
The real reason to check AI text is a different one anyway: invented prices, wrong opening hours or services you don't offer hurt you long before any authority does.
Fines – and why panic is still wrong
Yes, the AI Act provides for fines of up to €15 million or 3 percent of worldwide annual turnover for breaches of the transparency rules. That is the ceiling for corporations, not the rate for a café without a chatbot notice. For small businesses the realistic risk lies elsewhere: cease-and-desist letters from competitors, and lost trust when guests notice the photos aren't real.
In Austria, the RTR's AI Service Centre explains the new duties free of charge. Anyone trying to sell you an expensive "AI Act compliance" package for a simple website should be able to explain exactly which of the four points above even applies to you.
A ten-minute checklist
1. Is there a chatbot on your website? Then its first sentence must say it is an AI. 2. Do realistic AI images show your premises, team or products? Replace them with real photos or label them. 3. Do you publish AI text unchecked? From now on, read it before it goes live. 4. Does every fact an AI wrote match reality?
If you like, we'll look at your website for free and tell you in a short video whether any of these points apply to you.