Skip to content
KI-XellenzAI that knows everyday project work and takes the load off.
DE EN

Minutes and reports

AI texts people enjoy reading: minutes without filler

updated on 7 min read

This article describes the situation in Germany (German law and practice). It is a translation of the German original and is not legal or tax advice.

Martin Felux, civil engineer and industrial engineer

A text does not sound machine-made when your facts are in it: who does what by when, which figure applies and what was decided. For that you need a brief that names the reader and the purpose and includes your own notes. You then cut the AI draft hard and read it aloud. The AI words the text, you supply the content. Sources as of November 2026.

How to recognise machine-made minutes

The signs below are the author’s observations and not measurements. The sources at the end only show what makes texts hard to read in general. This article does not claim that programs can detect AI texts.

Typical is an introduction that announces the topic (“The most important results are summarised below”). Then come sentences of the pattern “not only A but also B” and lists that have three items out of habit, even when there is only one. Add unsupported judgements (“successful”, “comprehensive”, “constructive”), dashes used for seasoning and bold labels in front of every paragraph. The text ends with a sentence that says nothing: “Overall, the cooperation was positive.”

Some of these signs are also named in guides to public-sector language (the sources are German, so the examples refer to German style). In its 2002 handbook, the Federal Office of Administration advises against pushing verbs out with nouns. When something “comes to distribution”, the text bloats. With the passive, it often stays open who is meant to act: if something “has to take place”, the person responsible is missing. The same handbook demands that the same thing always be called by the same expression, because switching expressions to avoid repetition does not make the statement clearer. Client, employer and principal in the same minutes are such a case. The Bavarian Interior Ministry recommends stringing together no more than four nouns, because noun chains shorten the statement but make it harder to understand.

The Hohenheim comprehensibility index measures features of this kind, among them sentence length, share of passive sentences, word length and the share of abstract nouns. It runs from 0 (formally incomprehensible) to 20 (formally very comprehensible). The researchers name rambling sentences, abstract terms, compound words and unexplained technical terms as hurdles. They examined the speeches of chief executives at the 2018 annual general meetings, not site minutes. That the same hurdles apply in minutes is an assumption.

What your reader needs

The standard DIN ISO 24495-1 on plain language puts the target group first. According to the abstract at DIN Media, what counts is what readers want to know and need to know, and the context in which they use the document. They should find what they need and be able to understand and use it. The standard does not rest on readability formulas but on how successfully the target group can use the document. I have not read the text of the standard itself, which is paid.

That fits a qualification by the Hohenheim researchers: formal comprehensibility is not the only criterion, and content matters even more. For minutes this means answering the reader’s question. For the client it is usually: what must I decide, and by when? For the site manager: what is my task?

There are two limits. Terms from the contract, a statute or a standard stay unchanged, even when they sound clumsy. And minutes can later serve as evidence. A phrase such as “concerns raised” then has a meaning that your legal adviser assesses, not the AI.

From notes to a readable text
Text version of the graphic
  • 1. Collect the facts: Notes, figures, names and dates from you
  • 2. Write the brief: Reader, purpose, length, tone, two samples
  • 3. Draft the text: The AI words it from your notes (the AI prepares)
  • 4. Cut and check: Filler out, owners and deadlines in
  • 5. Read it aloud: Change what you would never say

Writing the brief

A request such as “Write minutes” produces generalities because the AI has nothing else to work with. Five details make the difference.

Name the reader and the purpose (“The client reads for two minutes and has to make a decision”). Supply the facts yourself, as notes with names, figures and dates, and demand that nothing is added that is not in them. The AI should mark gaps as questions, not fill them. Set the length and tone (“at most 120 words, factual, formal address”). Attach two earlier texts of yours that you consider good, with the instruction to adopt their wording and sentence length. And rule out what you do not want to read: no introduction, no closing sentence, no judgements, verbs instead of nouns, always name who acts.

Enter names, prices and contract data only into a tool that your company has approved for this.

Editing the draft

The draft is raw material. The same principle appears in the article A fixed workflow, one AI step, human approval: the AI proposes, a person approves. Four moves help when editing.

Cut first. Every sentence must give the reader something they do not yet have. Then make it concrete: where it says “soon”, put a date; where it says “delay”, put the number of days. Third, name a person responsible and a deadline for every task. A sentence with neither is usually an observation, not an agreement. Finally, read the text aloud. Where you stumble or say something you would never say, change it.

Three examples

All dates and roles in these examples are invented.

Site meeting minutes, before:

In the context of the site meeting of 12 October, the essential topics of the current construction progress were discussed comprehensively. It was established that the coordination of the trades is not only a central challenge but also requires continuous alignment. The delivery of the windows is expected to be delayed. A solution is to be found promptly. Overall, the meeting was constructive and goal-oriented.

After:

The windows arrive on 4 November instead of 21 October, according to the supplier. The site manager will clarify by Friday, 16 October, whether the drylining contractor can bring forward the work on the second floor. The window fitter will confirm a firm delivery date in writing by the same day.

Introduction, the “not only, but also” formula and closing sentence are gone. Date, people responsible and deadlines are new. The 56 words of the German original became 37, and the reader knows who does what.

Monthly report, before:

In the reporting period, pleasing progress was achieved in implementing the project. The shell construction work was successfully continued, the quality of execution is high and the cooperation of those involved is smooth. The cost situation remains stable.

After:

In October the structural contractor poured the slab over the first floor five days later than planned (14 instead of 9 October, because of rain). The schedule buffer is ten days, so the delay is absorbed. Invoiced to date are 1.84 million euros against a budget of 1.90 million euros for the shell. A variation of 12,000 euros for additional openings is outstanding and will be reviewed by 20 November.

“Pleasing”, “successful” and “stable” are replaced by figures. The reader can now judge for themselves.

Email to the client, before:

Dear Mr Muster, I hope this email finds you well. I would like to take the opportunity to inform you about the current status. As already discussed, there is unfortunately a delay, which however gives no cause for concern. Please do not hesitate to contact me with any questions.

After:

Dear Mr Muster, the windows arrive on 4 November instead of 21 October. The drylining contractor can bring forward the work on the second floor, which costs 2,400 euros more according to the quotation. Please decide by Monday, 19 October. I will call you on Thursday.

The email starts with the news and ends with a request including a deadline.

Phrases you can delete

Check each of these phrases individually, as not every one is wrong everywhere. The English equivalents of the German phrases:

  • “In the context of …” and “The following sets out …”
  • “It was established that …” (who established it?)
  • “not only …, but also …”
  • “promptly”, “shortly”, “in due course” without a date
  • “successful”, “comprehensive”, “goal-oriented”, “constructive” without evidence
  • “Overall, … went well” and “In summary, it can be said …”
  • “I hope this email finds you well”
  • “Please do not hesitate to contact me” (better: telephone number and time)

What you can do now

  1. Take one of your own minutes from last month and mark every sentence without a name, figure or date. Delete what is left if the reader will not miss it.
  2. Write a standard brief with reader, purpose, length, tone and the things that must not appear. Add two good texts of your own as samples.
  3. Write the next minutes following the steps in the graphic and time yourself. How to measure this properly is covered in Measuring time savings instead of promising them.
  4. Before sending, ask: after this text, would the recipient know what to do by when?

General information, not legal or tax advice. Your tax adviser, data protection officer or lawyer will clarify the individual case.

Frequently asked questions

Is it enough to tell the AI not to write filler?

The ban belongs in the brief, and the more concrete it is, the better (for example: no introduction, no closing sentence, no judgements). How much slips through anyway depends on the tool. Test it on your own texts and delete the rest yourself.

Can you recognise an AI text by such filler?

Filler is a nuisance in any text, whoever writes it. This article says nothing about whether programs can reliably detect AI texts. What matters more is that your reader understands who does what by when.

May I put minutes with names and prices into an AI tool?

Only into a tool that your company has approved for this and for which data processing has been settled. Your data protection adviser settles that, not this article.

Sources

  1. Universität Hohenheim: Klartext oder Kauderwelsch? Formale Verständlichkeit der CEO-Reden 2018 (Hohenheimer Index, Juni 2018) (University of Hohenheim: formal comprehensibility of CEO speeches, 2018) (retrieved on 13 November 2026)
  2. DIN ISO 24495-1:2024-03, Einfache Sprache, Teil 1: Grundsätze und Leitlinien (ISO 24495-1:2023), Seite und Kurzbeschreibung bei DIN Media (plain language standard, product page with abstract) (retrieved on 13 November 2026)
  3. Bundesverwaltungsamt: Arbeitshandbuch „Bürgernahe Verwaltungssprache“, 4. Auflage 2002 (Kopie der Universitäts- und Landesbibliothek Bonn) (German Federal Office of Administration: handbook on citizen-friendly administrative language) (retrieved on 13 November 2026)
  4. Bayerisches Staatsministerium des Innern: Freundlich, korrekt und klar, 4. Auflage, Stand August 2022 (Kopie der Universität Passau) (Bavarian Interior Ministry: guide to citizen-friendly language) (retrieved on 13 November 2026)

Read next