How Branddock marks AI content

The EU AI Act (Article 50) requires that what an AI system makes is marked in a machine-readable way as made by AI, and that the marking can be checked. Most of what Branddock makes with AI carries such a marking: images, video, audio, and text in files and on pages. What is not marked yet is listed below.

• Images and video: usually the maker of the model already puts Content Credentials (C2PA) in the file, signed, with the source type "made with AI". We leave those untouched, and they are then the marking. If they are missing, or disappear through an edit such as placing your logo, we add the IPTC source type (Digital Source Type) to the XMP metadata.
• Audio (voices and sound effects): the same XMP in the mp3's ID3 tag, plus a DigitalSourceType text field.
• Pages that Branddock hosts: schema.org digitalSourceType in the page's JSON-LD, and a meta tag.
• Downloads: a PDF in its XMP, an HTML file in its JSON-LD and a meta tag, markdown in its front matter, JSON in a provenance field, and a ZIP in provenance.json with the origin of each file.
• The API and the MCP server: a provenance field next to text that Branddock generated.

• Made with AI (trainedAlgorithmicMedia): Branddock made it with an AI model.
• A mix (compositeSynthetic): your own text or brand data, completed with parts made with AI.
• Edited with AI (compositeWithTrainedAlgorithmicMedia): an existing image that was edited with AI.

• Plain text, copied or as a .txt file: it cannot carry a machine-readable marking.
• Short texts, such as captions, alt texts and labels.
• Images in gif, avif or svg.
• Email that Branddock sends for you, social media posts and posts in WordPress: we do not mark those channels yet.

With the free tool Check AI marking you see what is in a file or on a page that Branddock hosts. The tool checks a file in your browser: it never leaves your device. If a file carries Content Credentials, the tool reports that, with the source type when it is readable in the file. The tool does not verify the signature; you can do that at contentcredentials.org/verify.

Our own marking sits in the metadata and is not signed. It can disappear, for example with a screenshot, when converting to another format or when uploading to a platform that strips metadata, and someone can imitate it. The Content Credentials from the maker of the model are signed, but they disappear in the same ways. We do not add an invisible watermark. A marking you find therefore says what the file says about itself, and is not proof.

Do not remove or change the marking on purpose. Article 6 of the terms of service says so too.

If you publish a deepfake, or AI-generated text that informs the public on a matter of public interest, the AI Act may require you to disclose that visibly (Article 50(4)). For such text this does not apply if a person has reviewed it editorially and someone holds editorial responsibility for it. That is your responsibility; this is not legal advice.