AI generated content has to be recognisable as AI. Here is what the requirements mean in practice, when they apply, and what the labelling can look like. Use this page to look things up when you are in doubt.
Most questions land in one of these four. Each one states the requirement, with an example beside it of what it can look like in practice.
If you make photorealistic material that people could genuinely mistake for real persons, objects or events, it has to be clearly marked to the recipient.
The label has to be visible where the image is shown, not buried in a footnote or in the terms and conditions. A small mark on the image plus an explanation nearby covers both at once.
If you use an AI driven chatbot, a virtual assistant or an automated phone system for customer service or sales, you are obliged to tell the customer.
The customer has to be told before the conversation starts, or at the very latest as it starts. It has to be plain to see in the chat window, so there is never any doubt that no human is at the other end.
When a member of staff approves the draft before it goes out, it is that person who answers. The reply then does not as a rule have to be labelled, provided the review is real.
Tools that generate content mark it themselves. The requirement for machine readable marking sits in the EU AI Act, article 50, and in practice it happens in two ways: as information inside the file, typically following the open standard C2PA, also known as Content Credentials, and as an invisible watermark woven into the image itself, for example Google's SynthID.
Neither can be seen with the naked eye. The difference is how well they hold: the watermark sits in the image and usually survives cropping and compression, while the information in the file can disappear when you compress, convert, screenshot or upload through a tool that strips metadata. Lose the C2PA stamp and you lose the formal documentation of where the file came from.
Woven into the image pixels, not into the file's information. It cannot be seen, and usually survives cropping and compression. Read with the provider's own tool.
Longer AI written texts meant to inform the public about matters of public interest have to be clearly declared.
Product copy, category copy and campaign copy normally fall outside. Once you move into health, finance, the environment or politics, it applies.
Labelling is about what the customer is told. The other question that always follows is what happens to your own material along the way. In short: it stays with you.
Access behind login with two-factor, or frictionless IP access from the office. You choose the level.
Write access is gated per integration via whitelist. Elke only acts where you allow it.
Your data is hosted in Amsterdam. We sign a data processing agreement with you.
Your sensitive data, product feeds and internal information are never used to train public AI models.
The AI infrastructure and the agents are secured through closed integrations, meaning secure APIs and MCPs. There is no open way into your systems.
Every exchange between the AI systems and platforms such as Google Ads, Meta and Klaviyo runs over approved, encrypted connections.
The rules are about transparency. The recipient has to be able to tell that they are dealing with AI. Three things decide whether a requirement lands on you.
The requirement sits with whoever puts the content in front of the recipient. That is you, not the tool the content was made in.
The closer the content comes to reality, the stricter the requirement. A photorealistic image of a product is a different matter from a sketch or an illustration.
No distinction is made between shop, newsletter, advertising and social media. The labelling has to travel with the content to wherever the customer meets it.
We produce and label AI content every day, and we follow the requirements closely as they take shape. This page is the practice we work by ourselves, and it is updated when the rules are. Use it as an overview of the rules that apply once content is made with AI.
Send it to us and we will look at it together, from a creative and technical perspective.