AI Content Disclosure

When you give a topic that is not in our built-in library, a language model may be asked to propose the words and write the clue lines. That is the whole of its role: vocabulary and wording.

• Choosing where each word sits in the grid.
• Filling the remaining squares with letters.
• Checking the result: every answer present, no answer readable in two places, every letter run in a crossword being a real answer, every scramble using exactly its answer's letters.
• Building the answer key.
• Laying out the page, the margins and the typography.

A model never draws a grid and never writes the answer key. Text that has to be exactly right — the words, the clues, the title, the page footer — is typeset by our code as real text, never drawn as an image.

Language models produce confident-sounding errors. A clue may be imprecise, outdated, or simply incorrect. Gridwyn therefore shows every word and clue in an editable table before you export, and asks you to read them. On a page destined for an exam or an assessment, please check every line.

Every PDF and PNG carries provenance metadata declaring AI-assisted content. In PNG files it is written as tEXt and iTXt chunks including an XMP packet; in PDF files it is written into the document properties. The digital source type we declare is the IPTC term:

http://cv.iptc.org/newscodes/digitalsourcetype/compositeWithTrainedAlgorithmicMedia

That term means media combining software-composed elements with output from a generative model — which is exactly what a Gridwyn page is. The marker travels with the file when you email it, upload it or hand it on. In addition, every page prints a visible line in its footer saying the same thing in plain words.

This satisfies the transparency duty in Article 50 of the EU AI Act for AI-assisted output, and it is there whether or not you are in the EU.

The photographic images and the short video clips on this site were generated with AI image and video models. They are illustrations of desks, classrooms and printers — no real person, place or event is depicted, and none of them contains text. Every puzzle page shown on this site, by contrast, is a genuinely generated page, not a mock-up.

Topics, word lists and clues you enter are not used to train any model, ours or a provider's. See the Privacy Policy for retention periods. If you have a question about any of this, write to support@adeltahar.shop.