Four JSON files live under /daten/: cases, laws, sources and decision tree tree.v1. Every page has a Markdown version at /path/index.md, plus llms.txt, llms-full.txt and an Atom feed. Everything is licensed under CC BY 4.0 with attribution to “ai-label.app”.
Datasets under /daten/
Each file is a JSON object that starts with the same fields: license and licenseUrl (CC BY 4.0), publisher, source, updated (ISO date) and note (a reminder that this is not legal advice). The content follows.
| File | Content |
|---|---|
faelle.json | Typical image cases from online stores such as an AI model, an AI background or a cut-out, each with the answers in the decision tree, the result code (table below) and the paragraphs of the guidelines it rests on. As pages under Cases. |
gesetze.json | AI labelling rules worldwide, for each country or state with legal basis, obligation, start date, sanctions and source. Readable under Worldwide. |
quellen.json | All sources of the website in the field sources, each with id, titel (German and English), von (publisher), url, datum (date) and art (type: gesetz, leitlinie, kodex, behoerde, verband, norm, plattform). Readable under Sources. |
generatoren.json | AI image generators such as ChatGPT, Midjourney or Firefly, each with the documented state of C2PA, IPTC field, invisible watermark, PNG text fields and what is left after upload, with sources. not_documented means no source was found, not that the marking is absent. Readable under AI generators. |
fragebaum.json | Decision tree tree.v1 of the labelling check with questions, allowed answers, texts in German and English and the possible results, identical to the one in the Shopify app. |
Result codes in cases and decision tree:
| Code | Meaning |
|---|---|
A | Labelling required |
AK | Required, lighter form |
C | Labelling recommended |
CF | Third-party image, recommended |
D | Legacy image |
B | Not required under the guidelines |
B0 | Not an AI image |
N | No indication of AI |
Example request:
curl https://www.ai-label.app/daten/faelle.jsonCORS is open, so the files can also be loaded from the browser on other websites. File and field names are German; text fields are bilingual where marked.
The datasets are read-only. An API that labels images or writes metadata is planned; until then, the tools do this in your browser.
Markdown version of every page
Add index.md to a page address, for example /en/glossary/index.md, to get the page as Markdown with title, address, last-updated date, short answer, content, questions and sources. It matches the HTML page and is linked in every page's head as rel="alternate".
llms.txt and llms-full.txt
/llms.txt is an overview for AI assistants with key facts, citations and all pages with a short description. /llms-full.txt contains all texts of the website in one file.
Feed
Changes to content and the legal situation are recorded in the changelog, as an Atom feed at /feed.xml (entries in German).
Widget for websites
One line of code labels AI images on your own website. The widget page explains embedding and marking; guides per store system are under Integrations.
Licence
Datasets and texts are licensed under CC BY 4.0. You may share and adapt them, including commercially, as long as you credit “ai-label.app” with a link to the source. Please include the last-updated date when you use cases or laws, because the legal situation changes.
The EU icons are excluded; the European Commission's terms apply to them.
The data explains the law; it is not legal advice. A result from the cases or the decision tree does not replace a decision on the individual case.
