How we work with AI

This page says where AI does part of our work and where a human decides. We write it down because it is exactly what we sell: the machine takes the monotonous work, people make the decisions. A sentence nobody can check is just a slogan.

On Mondays an agent reads our sources from the past week and proposes three topics. A human picks one. Then an agent writes the German draft, and another reads it against our own rules and reports what it takes issue with.

After that it is a human's turn: read, rewrite, approve. No post goes online without that approval. This is not a house rule but a setting in the database: the credentials the agents work with may create a draft and may not publish one.

Once a post is out, an agent compares the draft against what is online. Whatever a human changed is stored as an example, and the next draft learns from it.

Every machine-drafted post carries a line under its headline saying which of our robots wrote the draft and which human edited and approved it. A post a human wrote from start to finish carries no such line.

The English versions are machine translations of the German texts, checked and approved by a human. That is on the post as well.

The robots on this site and the cover images of the posts are made with an image model. The mascot and its crew come from a reference sheet of our own, so they stay the same characters across the site.

What we do not do: photorealistic pictures of people, places or events that never existed that way. Our pictures are recognisable as drawings. Should a photorealistic one ever appear, we will label it on the picture itself.

This site was built with an AI assistant too, the text as much as the code. Every page was read, changed and approved by a human; the imprint and the privacy policy were reviewed by the owner.

There is one place on this website where it works differently, and only one. Whoever describes their work on the page „Was eine Maschine bei euch übernehmen könnte“ receives an assessment written by an AI system that nobody here has read beforehand.

That is why the notice about it stands above the input field rather than below it, and why the assessment is expressly not binding. It names no prices, no durations and no promise that we build anything. That is not a form of words, it is built: the shape of the answer has no field a number fits into.

Two rules hold the answer to what you actually wrote. Every point it makes has to quote the passage of your text it rests on, and a quote that cannot be found there is removed from the answer by machine before you see it. And every point carries a sentence about what stays with a person.

The transparency duties of the European AI Act have applied since 2 August 2026. Anybody who interacts directly with an AI system has to be told so. On the possibilities page that notice therefore stands above the input field.

For texts that inform the public on matters of public interest a labelling duty is added. Exempt are texts that a person has reviewed in substance and for which a person holds editorial responsibility. Every one of our posts is reviewed and answered for by a person.

We write on every post how it came about anyway. A label alone only says whether a machine was involved, not who answers for the result. The second half is the one that matters to us.

Responsible for the content is the owner named in the imprint.

If you want to know how a particular post came about, ask: info@kimaschine.de. For every post we can say which agent worked on it and what a human changed.

Last updated: August 2026