NADOVO Plattform
AI-generated You don’t have to label everything your AI produces. But you must be able to prove why not.
That is the uncomfortable truth behind the transparency obligations of the EU AI Act, which apply from 2 August 2026. An AI text that a human has reviewed needs no label. But if someone asks a year from now whether that review really took place, it is claim against accusation. Unless you have evidence.
That is exactly what the AI Content Register is for, new in the beta of the NADOVO platform.
Two kinds of evidence, one register
Article 50 gives companies two different tasks, and both end at the same question: can you prove it?
For AI texts on matters of public interest, the label is waived if a human has reviewed the text and someone holds editorial responsibility. This exception is valuable, but it wants to be proven: who, when, what. The AI Content Register records exactly that, with person and timestamp.
For images, videos and chatbots there is no such exception. Photorealistic AI images count as deepfakes under the draft Commission guidelines and must be labeled. Here the register proves something else: that you labeled, how and since when. And for borderline cases, why you didn’t have to.
The difference: your evidence knows its origin
A note in a folder can document things too. What it cannot do: prove what the content was created with.
Every entry in the AI Content Register is linked to the AI process that produced the content, that is, the system, the use case and the risk class from your AI inventory. Which tool, which model, used for what, who reviewed it: one click, one picture. And everything runs into the platform’s immutable audit trail, retained for ten years, not editable after the fact. Evidence you can edit is not evidence. When it matters, the platform delivers the proof including an audit trail extract as a protected PDF report, at the push of a button.
Captured in seconds, not in forms
Documentation rarely fails on will, mostly on effort. That’s why capture is radically lean: three required fields, one form, done.
It gets even faster with the metadata prompt. You copy it in the capture dialog for the respective AI process, attach it to your request in the AI tool, and the tool delivers the metadata for the register entry along with the content. Upload, quick check, save. This works with any tool, without a plugin and without an API.
One thing the upload deliberately does not take over: the review. Release and labeling decision remain a human step in the platform. An AI certifying itself as reviewed would be worthless in a dispute. This separation is what makes the evidence credible.
The release thinks along
The release step is also the moment the platform explains the legal situation, matched to the content type.
For a text, the register first asks whether it covers a matter of public interest at all. If not, there is no labeling obligation, and exactly this assessment is documented. If it does, you confirm review and editorial responsibility, and the register tells you: the exception applies, no label needed, this entry is your proof. For a video it asks: could it appear real? If yes, it tells you clearly that it must be labeled, and what the label has to look like. A chatbot is always made recognizable.
So you don’t need a legal opinion next to your keyboard. The decision stays with you, but the rules are right where you need them: at the moment of decision.
What the register is not
Plain words, because this matters to us: there is no legal obligation to keep such a register. The AI Content Register is evidence provisioning, not a mandatory program. And it is no free pass: what must be labeled must be labeled, the register documents the clean implementation.
But exactly this provisioning is what separates, when it counts, the companies that can prove from those that must protest. Article 50 carries fines, and the question “Is this AI, and where does it say you reviewed it?” comes not only from authorities, but also from competitors, customers and journalists.
Doubly interesting for agencies
Anyone producing AI content for clients must be able to clear themselves twice: towards the law and towards the customer. With the multi-tenancy of the platform you keep the AI Content Register per client, with clean evidence for each customer.
And because the register has just entered beta: we are continuously looking for beta testers, whether you produce AI content yourself as a company or want to manage the register for your clients as an external AI compliance partner or consultant. Just get in touch via our platform page.
One question to finish. If tomorrow someone points at an AI text from last quarter and asks who reviewed it: could you prove it? If not, that is exactly the gap this register closes. What the EU AI Act specifically requires for labeling is covered in our article on AI content labeling, and how to get started in general is what we clarify in AI compliance consulting. Where you stand today is shown by our self-check.
About the author
Jochen Stier is a co-founder of NADOVO with over 20 years of experience in process management and IT service management. He helps German SMEs implement the requirements of the EU AI Act systematically and pragmatically. His 5-phase NADOVO framework combines regulatory requirements with practical feasibility, without enterprise budgets or complex tools.
Further reading:
- Code of Practice on labeling AI-generated content (European Commission)
- EU AI Act, Article 50 full text (EUR-Lex)