Since 2 August, your visitor must know they are talking to AI
The transparency obligations of Article 50 of the EU AI Act have applied since 2 August 2026. The European Commission’s guidelines state that a general reference in your terms of service is in principle not sufficient. We map which obligation rests on you and which on your vendor, arrange the disclosure, and deliver an evidence file.
3 working days · fixed price · credited against follow-up work
What changed
Article 50 of the AI Act contains four transparency obligations. They apply from 2 August 2026.
| What | Who | When |
|---|---|---|
| Disclose that a person is interacting with an AI system | Provider | Now |
| Machine-readable marking of AI-generated audio, image, video and text | Provider | Now — existing systems until 2 December 2026 |
| Inform people subject to emotion recognition or biometric categorisation | Deployer | Now |
| Visibly label deepfakes and AI text informing the public on matters of public interest | Deployer | Now |
Paragraph 5 governs the form: the disclosure must be clear, distinguishable from other information, provided at the latest at the first interaction, and meet the applicable accessibility requirements.
The Commission’s guidelines — non-binding, but the best available guidance — state that a general reference in the terms of service or somewhere on the website is in principle not sufficient; it should be visible during the interaction itself, as a label, banner or badge.
Where the obligation sits
Paragraphs 1 and 2 address the provider of the AI system, paragraphs 3 and 4 the deployer. If you buy a chatbot from a vendor, you are formally not the addressee of paragraph 1 — but your contract decides who configures the disclosure and who maintains it. That division of roles is exactly what we document for you.
The maximum fine for breaching Article 50 is 15 million euro or 3% of worldwide annual turnover; a proportionate approach is foreseen for SMEs (sources differ on how that works out exactly — ask a lawyer).
What we do
We audit
- Every touchpoint on your site where a visitor talks to AI — chat widget, voice bot, form assistant
- Whether the disclosure is visible at the first interaction, and not only in your terms
- Whether AI-generated images and video on your site are labelled
- Whether your AI text falls under the marking obligation — usually it does not, and we write that down
- What your vendor handles and what you must handle yourself — including the exact question to put to your chatbot vendor
You receive
- A findings report per touchpoint, with a screenshot and the exact text that should be there
- Ready-to-paste disclosure copy in Dutch, English and German
- An evidence file: screenshots, configuration export, vendor information. Article 50 has no formal conformity assessment, but this is what you show when someone asks
- A list of what you do NOT need to do — this usually saves more than the check costs
What we are not
We are not a law firm and not a compliance consultancy. We audit against the four transparency obligations of Article 50 plus paragraph 5 on the form of the disclosure, and we document what is there and what should be there. For a legal assessment of your specific situation we refer you to a lawyer. That boundary is stated in the report.
Frequently asked
We only use a third-party chatbot. Does this apply to us?
Probably less than you think — which is exactly why the check is worth doing. The obligation to disclose that it is AI (paragraph 1) formally addresses the provider of the system, so your vendor. Sources differ on how far the deployer’s responsibility reaches. What is certain: it is your website, your customer, and your contract decides who arranges it. We deliver the exact question to put to your vendor, the text that should be there, and the record that it was arranged.
We only use ChatGPT internally. Does this apply to us?
For internal use, usually not. The obligations concern systems that communicate directly with people and content that you publish. An internal memo does not fall under them. That is stated in the report as well.
Do we have to label all our AI text?
No — this is the most common misconception. The marking obligation for text applies to deepfakes and to text informing the public on matters of public interest, unless it has been reviewed under human editorial responsibility. For most business content it does not apply.
Do you guarantee I will appear in ChatGPT?
No. Nobody can. We make sure you are not skipped for technical reasons — a different and solvable problem.
Is this legal advice?
No. It is a technical audit against a published set of obligations, plus documentation. For a legal assessment of your specific role under the Act, consult a lawyer.
The check costs less than one misunderstanding.
Three working days after you send the URLs you know exactly what is on your site, what should be there, and what you can ignore. Fixed price, no open ends.