Skip to main content

Tasks that show students' own thinking

AI solves many traditional tasks in seconds. With different tasks, what students think for themselves becomes visible again.

6 min read

Two teenagers working together on a laptop

Key points

  • Programs that are meant to detect AI-written text are unreliable and no basis for a mark.
  • Tasks that make the thinking process visible work better than bans.
  • How much AI is allowed is set in advance for each task. Then the same rules apply to everyone.

AI solves many traditional tasks in seconds: a summary, an argumentative essay, a set of maths problems. According to the JIM study 2025, a German survey of young people's media use, 84 per cent of 12 to 19-year-olds have already used ChatGPT. So how can teachers still tell what someone has thought for themselves?

Through the task. Students' own thinking becomes visible when the task shows the thinking process and it is clear beforehand how much AI is allowed. Trying to find out afterwards whether AI was involved does not work.

Why detection is not enough

Programs that are meant to detect AI-written text are often wrong, in both directions. They mistake texts that students wrote themselves for AI text and miss texts that were produced with AI. OpenAI discontinued its own detection tool in July 2023 because of its low accuracy.

A suspicion alone is therefore no basis for a mark. A conversation about how the student reached the answer is better. How to deal formally with a suspicion of cheating is set out in the rules of your state and your school.

Four formats that show students' own thinking

In October 2024, the Bildungsministerkonferenz, the conference of education ministers of the German states, recommended shifting assessment further from the product to the process. Reflection by learners plays a decisive role in this. New assessment formats are also to include work prepared over a longer period, collaborative and dialogue-based work, and presentations. Where AI is used for a product, the presentation and especially the defence should take into account skilled use of AI and the ability to reflect on the results.

Four formats for the classroom follow from this:

  • Process, not product: Interim stages, a learning journal, drafts with reasons.
  • Personal connection: Students' own observations and experiences from where they live or from lessons.
  • Checking an AI answer: Examine an AI solution for errors and improve it.
  • Oral elements: Short conversations about the student's own approach.

Four tasks, before and after

In all four examples, the learning objective stays the same. What is new is that the route to it becomes visible.

German, grade 7. Before: “Write a summary of Goethe's ‘The Sorcerer's Apprentice’.” After: “The AI has written a summary. Find three errors and back each one up with evidence from the text.” To find the errors, students have to know the ballad and read it closely. The teacher creates the AI summary in advance.

Maths, grade 7. Before: “Solve ten percentage problems.” After: “An AI has solved three problems, and one of the solutions is wrong. Find it, explain the mistake in two sentences and write a problem of your own in which it is easy to make the same mistake.” The student's own problem shows whether they have understood the mistake.

History, grade 9. Before: “Write an essay on the causes of the First World War.” After: “Ask an AI to name three causes. Check each one against your textbook, add one that is missing and explain which you think is the most important. In the next lesson, you will explain your choice in three minutes.” Here, AI is allowed for ideas. What is assessed is the checking, the reasoning and the conversation.

Geography, grade 6. Before: “Describe how a river shapes a landscape.” After: “Go to a stream or river near you. Note down three things you see there and use your textbook to explain how they came about. You may include a sketch.” No AI can provide the student's own observation. The explanation shows whether they have understood how it all fits together.

When learners work with AI themselves, they use a tool that the school has approved and that is permitted for their age. Personal data does not belong in any prompt.

How much AI is allowed?

Decide for each task how much AI is allowed. A well-known scale for this is the AI Assessment Scale by Perkins, Furze, Roe and MacVaugh (2024). Its five levels range from “no AI” through “AI for ideas and outlining”, “AI for revising” and “AI completes parts of the task, with disclosure” to “full AI use”. It was developed for higher education and transfers well to schools.

Which level fits depends on the learning objective:

  • If a basic skill is being practised or tested, such as spelling or mental arithmetic, then “no AI” applies and the task is done in class.
  • If the focus is on ideas and structure and the text is to be the student's own work, then “AI for ideas and outlining” fits.
  • If the work is done at home, then build in interim stages or a short conversation, because a ban can hardly be enforced there.
  • If working with AI is itself the learning objective, such as writing good prompts and checking results, then “full AI use” with a written reflection fits.

Write the level on the worksheet. A header you can copy:

  • AI on this task: Level 2, AI for ideas and outlining.
  • What you do yourself: The text, the evidence and the reasoning.
  • At the end, state: Which AI you used, what for and which parts are your own. For example: “Outline with AI, text written by me.”

An assessment grid for students' own work

If the process counts, assessment needs criteria for the process. The following grid can be applied to most tasks. Rate each criterion as “evident”, “partly evident” or “not evident”. You or your subject department decide how much weight each criterion carries in the mark.

  • Approach: Interim stages or notes show how the result came about.
  • Disclosure: Tool, purpose and the student's own contribution are fully stated.
  • Checking: AI results were checked, errors named or details compared with the textbook and sources.
  • Reasoning: The student's own decisions are justified in a way others can follow.
  • Explaining: The student can explain the result in their own words.

The teacher always gives the mark. AI can prepare feedback; it should not do the assessing.

A conversation about the approach shows what has been understood.
A conversation about the approach shows what has been understood.

When a suspicion arises

Sometimes a piece of work does not match what you know of a student. Instead of a detection tool, a calm conversation of a few minutes helps. These questions are suitable:

  • How did you go about it? What did you start with?
  • Which part was the hardest, and how did you solve it?
  • Explain this term from your text to me in your own words.
  • What would you change if the task were set slightly differently?
  • Did you use AI? If so, what for?

Afterwards, note down in a few sentences what was discussed. Anyone who has worked something out themselves can usually explain it too. If not, that is a reason for a new agreement.

Next steps

  • Choose a homework task from your lessons that AI could complete entirely today, and redesign it following one of the examples.
  • Add the header with the permitted level to your next worksheets.
  • Agree the grid in your subject department, so that learners in all classes know the same criteria.
  • Decide how learners disclose AI help. The article AI in schools in Hesse describes a standard format for the whole school.

If your teaching staff want to practise this together: in the accredited AI training for teachers from lindenbrook, “Rethinking tasks” is one of four parts. Your staff work on homework tasks and formats from their own lessons.

Which task from your lessons will you redesign first?

Frequently asked questions

Should we ban AI for homework?

A ban can hardly be enforced. The conference of education ministers of the German states (Bildungsministerkonferenz) also considers a general ban on work done at home to be ineffective. Tasks in which the student's own approach counts work better, together with agreements made in advance on when AI is allowed.

Who gives the mark?

Always the teacher. AI can prepare feedback; it should not do the assessing. The conference of education ministers of the German states also sees assessing performance as a task carried out solely by teachers.

Can a detection tool prove a suspicion?

No. Such programs are often wrong, in both directions. OpenAI discontinued its own detection tool in July 2023 because of its low accuracy. A conversation about the approach tells you more.

How do learners disclose AI help?

With three details: which AI they used, what for and which parts are their own. A sentence such as “Outline with AI, text written by me.” is often enough. Ideally, the school sets a standard format.

Isn't that a lot of extra work?

Redesigning a task takes time the first time. Many redesigned tasks can then be reused with every year group. If the subject department shares tasks and grids, the work is spread across the team.

Sources

  1. Perkins, Furze, Roe, MacVaugh: The AI Assessment Scale (2024)
  2. OpenAI: New AI classifier for indicating AI-written text (with a note on its discontinuation in July 2023)
  3. Weber-Wulff et al.: Testing of Detection Tools for AI-Generated Text (2023)
  4. KMK: Recommendation for education administrations on dealing with artificial intelligence in school education, resolution of 10 October 2024, PDF (in German)
  5. mpfs, JIM study 2025 (in German)

Updated: . This article is not legal advice.

AI training for teachersAccredited by the Hessian Teacher Academy · on site or online
See the training

Read next

Questions about your situation?

In a first call we discuss your questions and possible next steps. 30 minutes, no obligation, by phone or video.