How we write, and what we use AI for
MateMatic writes about artificial intelligence and uses it too. We say plainly what it helps us with, who is responsible for each text, and how we check before anything is published.
This policy covers news, Knowledge Base reviews, Academy guides, user guides and product pages on this site.
01Who is responsible
Wiesław Mazur, MateMatic's chief architect, is responsible for every text published on this site. A text goes live only after he decides it should. No tool publishes anything here on its own.
02What AI helps us with
We use language models when writing, mainly Claude by Anthropic. They help with drafts, with ordering an argument, with checking facts against sources and with editing. We do not sign texts with the model's name, because a person is responsible for the content, but we do not pretend we wrote without AI.
03How we check before publication
- Facts at the source. We check provisions, judgments and dates in the source text: in the Journal of Laws, in EUR-Lex, in Repertorium. We give the sources under the text, with a link or a document identifier.
- Editorial review. A separate reviewer, an AI tool running on our rules, points out weak spots: claims without support, vague generalities, overblown conclusions. The author fixes them; the reviewer does not write for the author.
- Mechanical checks. Typography, Polish spelling under the rules in force since 1 January 2026, empty buzzwords and text length. Knowledge Base reviews carry a legal disclaimer in the byline, and the site will not accept a review without one.
04What AI does not do here
- It does not publish. A person decides on publication.
- When it helps write texts for this site, it gets no data from the files of real cases. Professional secrecy applies to our tools too: public texts contain no client data, not even without a name, not even as a bare number or date.
- It is not a source. A statement about the law rests on a document, not on a model's memory.
05Why we write this, although we do not have to
Article 50(4) of the AI Act requires a deployer to disclose that text published to inform the public on matters of public interest was generated by AI. The same provision exempts content that has undergone human review or editorial control, where a natural or legal person holds editorial responsibility for its publication. In our assessment, our texts fall within that exception.
We say it anyway. We teach law firms transparency in working with AI, so we start with ourselves.
06Errors
Spotted an error in a text or in a source? Write to [email protected]. We will correct it in the text.