The key points at a glance
- Chatbots must make it clear that you’re talking to an AI, unless that’s already obvious.
- Deepfakes — realistic AI images, videos or voices of real people, places or events — must be labelled.
- AI texts on matters of public interest must be declared, unless a human has reviewed them and takes editorial responsibility.
- Swiss companies are affected once their AI content is used in the EU. There is no dedicated Swiss AI law yet.
What the EU AI Act requires
The EU’s AI Regulation sets out transparency rules in Article 50. These obligations have applied since 2 August 2026:
- Interacting with AI: Anyone offering a system that people interact with directly, such as a chatbot on a website, must design it so users know they’re talking to an AI.
- Deepfakes: Anyone using AI images, audio or video that closely resemble real people, objects or events must disclose that they are artificially generated or altered. Relaxations apply to works that are clearly artistic or satirical.
- Texts of public interest: AI-generated texts intended to inform the public must be labelled, unless they have been reviewed by a human who takes editorial responsibility.
- Machine-readable marking: Providers of AI generators must technically mark their output. This affects the makers of the tools, not you as a user. For systems already on the market before August 2026, a transitional period applies until 2 December 2026.
Why this affects Swiss companies: The regulation also applies to providers and operators outside the EU when the output of their AI is used in the EU. A chatbot on a Swiss website that also serves visitors from Germany or France falls under this.
And in Switzerland?
In February 2025, the Federal Council decided not to create its own, comprehensive AI law. Instead, Switzerland intends to implement the Council of Europe’s AI Convention and adapt existing laws where needed. A consultation draft has been announced for the end of 2026.
The Swiss FADP already applies to AI too, as the Federal Data Protection and Information Commissioner (FDPIC) has clarified. Anyone using AI must inform people transparently about its purpose, how it works and its data sources. Read more in the article AI and data protection.
What do I need to label?
| Content | Example | Label it |
|---|---|---|
| Chatbot | An AI assistant answers questions on the website | yes |
| Realistic AI photo | An AI image shows a “team” or “customers” that don’t actually exist | yes |
| AI voice, AI video | An explainer video with an artificial narrator | yes |
| Illustrations | Recognisably artificial graphics, icons, still life images | optional |
| Reviewed texts | Drafted with AI, revised and taken responsibility for by you | optional |
| Unreviewed news | Automatically generated reports with no human check | yes |
Our recommendation, even where it’s not required: be open about how you use AI. Customers forgive AI images, but not the feeling of having been deceived.
What an AI declaration looks like
The simplest approach has two levels: a short note directly on the content, and a page or section that explains your approach to AI.
- On a chatbot: “Hello, I’m the digital assistant for Muster AG. I’m an AI and can make mistakes. For binding information, please reach our team at …”
- On images: In the caption or alt text: “AI-generated illustration”.
- In general, for instance in the footer or legal notice: “We use AI tools when drafting texts and images. All content is reviewed and taken responsibility for by our team. We label realistic AI images.”
Transparency here is not a risk, but a trust advantage.
This article summarises the legal situation at the time of publication and does not replace legal advice for individual cases.
Using AI sensibly and transparently
Planning a chatbot or AI-generated content for your website? We’ll advise you on implementation, disclosure and data protection: book a free intro call.
