Motivation
The problem.
Many authors write with help from large language models. They use the machine for research, for polish, for brainstorming, sometimes for a passage. Most of them hide it. Saying so can cost a contract, a prize, a reputation.
So nobody knows how much LLM-assisted writing is already in print. The number is dark, and suspicion fills the gap. Readers turn into investigators. Writers turn into suspects. Detection tools promise to settle it and don’t — they flag the wrong people, especially writers whose prose sits outside the average.
Under those conditions, lying wins. The honest author is punished. The careful liar gets away. The reader can’t trust either signal.
The solution.
Maschinenschrift moves the disclosure to the author. The author says, in their own hand, where the machine helped and how much.
A standard, not a verdict. No purity stamp. No quality claim. No human-only badge. Voluntary. No registration. A book without a seal is not accused; a book with one is not certified.
The question changes shape. Not did she use AI? — but what does the work itself say about how it was made?
The seal.
Three axes. The author marks each one.
- Wortlaut — wording; how much of the surface (sentences, word choice, rhythm) came from the machine.
- Geist — how much of the substance (ideas, argument, structure) came from the machine.
- Beleg — the record; whether the work stands in the bibliographic world: findable, datable, identifiable.
Two rows of dots, zero to five per axis, count the machine’s share. A half-stroke below means correspondence is still pending. The author sets the seal and places it near the byline — cover, title page, about page.
That is the whole notation.