AI transparency

Five places, and no more. All five run on language models from Anthropic, which is our processor in this area. The contractual basis is set out in our data processing agreement.

Our articles and glossary entries come about the way most other text does today: a person writes, researches and edits, and language tools are involved along the way. Every article has a named author with a title, a photograph and contact details. That person has read the text through and vouches for it.

That is why we do not put an AI label on every single article. The disclosure duty for AI-generated text in Article 50 applies to text published in order to inform the public on matters of public interest, and it falls away where the content has been under human editorial control and a natural or legal person holds editorial responsibility. Ejerblik ApS does so through its named author.

We also believe a generic AI label would obscure more than it discloses. What matters to a reader is not whether a language tool was involved. It is who checked the figures, and who they can write to if the figures are wrong. Both are stated under every article.

We will label text explicitly if we one day publish something where a person has not reviewed the content before publication. We have not done so to date.

The images on ejerblik.com are licensed stock photographs and our own photography. They are not AI-generated, and there are no images on the site showing people, places or events that do not exist.

Video and speech, on the other hand, are synthetic: the explainer video is a digital version of Ejerblik’s founder, Jacob Lysholt, rendered from a script he wrote himself. Four rules apply to all AI-generated material:

• Metadata is kept. We do not strip the C2PA manifest or the EXIF fields that say the image is machine-generated.
• Visible labelling. The material gets a line that says so in words. For a video it sits there before you press play, so it also works for a reader who does not inspect the file’s metadata.
• No imitation of real people without consent. The only real person we reproduce digitally is Jacob Lysholt, with his own consent and only for text he wrote himself. We do not generate images or video of other named people, and we do not use AI to manipulate photographs of them.
• A person writes the script. Everything a digital version says has been written and approved by a person at Ejerblik before it is voiced.

Our AI assistant AInette also appears as the presenter in short walkthroughs of the platform’s features. She is a synthetic figure, not an imitation of any particular person, and she says in her first sentence that she is an AI. It also says so next to the video and in the picture. Short clips from the explainer and from AInette’s walkthroughs are also used in adverts on Facebook and Instagram. There the label stays in the picture throughout, because an advert is shown without the page around it and can be watched from the middle. If we start using other synthetic material, we will update this page before it goes live.

This list matters just as much as the one above:

This page is written to Article 50 of the EU Artificial Intelligence Act, whose transparency obligations have applied since 2 August 2026. The article sets four requirements that are relevant to a company like ours: that you are told when you are talking to an AI, that synthetic content can be identified, that deepfakes are declared, and that AI-generated text on matters of public interest is disclosed, unless a person holds the editorial responsibility.

In Denmark, the Danish Agency for Digital Government coordinates supervision of the regulation. On top of that, the general prohibition on misleading practices in the Danish Marketing Practices Act applies: we do not say that something was written or assessed by a person if it was not.

We are a small company, and the rules in this area are new. If you think something here is imprecise or missing, we would rather hear it from you than from a supervisory authority.