Creating Teaching Materials With AI: A Guide for Teachers

Creating teaching materials with AI means letting a language model draft worksheets, tasks, quiz questions and differentiated versions — while the subject-matter responsibility and the fact-checking stay entirely with the teacher. Used well, it saves noticeable time; copied in unchecked, it carries errors straight into the classroom.

What does "creating teaching materials with AI" mean?

This is not a button that spits out finished lessons, but a tool for the first draft. A large language model — ChatGPT, Gemini or Claude, say — produces task prompts, example texts, question sets or simplified explanations on request. The teacher then checks, trims, corrects and adapts them to their own class. Put differently: the AI is a co-pilot, not an autopilot. It takes over the mechanical wording, not the pedagogical decision.

This way of working has long arrived in schools. A RAND study on the 2023/2024 school year found that about a quarter (25 percent) of surveyed U.S. teachers used AI tools for instructional planning or teaching (Kaufman et al., 2025). In Germany, the Deutsches Schulbarometer by the Robert Bosch Stiftung shows a similar picture: among teachers who use AI regularly, it serves above all to create teaching tasks (58 percent) and to support lesson planning (56 percent). At the same time, 62 percent of respondents felt unsure or very unsure about using such tools — a sign that the gap is less about the technology than about clear routines.

Which materials can AI create?

For the recurring text building blocks of everyday school life, a language model is especially useful. Realistically, it can take off your hands, among other things:

• Worksheets and tasks with a model solution — from a maths sheet to a text analysis.
• Quiz and multiple-choice questions on a topic or a specific text, including plausible wrong answer options.
• Cloze texts, vocabulary lists and matching exercises in seconds rather than minutes.
• Simplified or shortened versions of a subject text for weaker readers.
• Summaries and key sentences you reuse as a board note or handout.
• Drafts of parent letters, feedback or lesson grids that you only need to adjust.

The common denominator: these are rough drafts that save you the blank page. How to fold such building blocks into a lean weekly prep is shown in the piece on preparing lessons efficiently. More ideas and tools for the school day are gathered on our page for teachers.

How do I write a good prompt for worksheets?

The quality of the material depends almost entirely on the instruction. A good prompt — the name for the input you give the language model — states context and goal as precisely as possible. It has proven useful to include these points:

1. Role and subject: "You are a Grade 8 biology teacher."
2. Learning goal: What should the students be able to do afterwards?
3. Format and length: for example a worksheet with five tasks on one A4 page, with solutions at the end.
4. Level: grade, prior knowledge, the difficulty tiers you want.
5. Frame: link to the curriculum, content you want or want to exclude.

The more specific the brief, the less you have to rebuild afterwards. A second round is part of the normal process: "make task 3 simpler" or "add a transfer task" often delivers the right variant in seconds. Anyone who saves a few good base prompts has a reusable template for the next prep.

How do I create good quiz questions with AI?

Quiz questions are a prime example, because a language model easily supplies ten variants on a topic. For multiple choice it pays to explicitly ask for good distractors — wrong answers that sound plausible on their own and capture typical misconceptions. Otherwise the AI happily produces one obviously correct option and three absurd ones. Also state whether you want pure recall or application and transfer questions.

There is a methodological bonus: regular self-testing is itself an effective learning method. The so-called testing effect describes how active retrieval anchors material more durably than mere rereading. Well-made quiz questions are therefore not just a check but part of the learning — provided every question and every answer marked correct has been verified.

Why is fact-checking AI mandatory?

This is the most important point. A language model predicts the most probable next word — it does not check whether a statement is true. Experts call a wrong but convincingly worded result an AI hallucination. A widely cited research review describes large language models as prone to generating "plausible yet nonfactual content" (Huang et al., 2023).

For teaching this means, concretely: dates, formulas, quotations, references, chemical equations or historical details can be wrong — and in a tone that signals no doubt whatsoever. Especially treacherous are invented references, quotes attributed to the wrong author, and seemingly clean calculation paths that still lead to the wrong answer. A model writes nonsense with the same confidence as truth.

So without exception: no AI material goes into the classroom unchecked. In practice that means three moves: compare facts against a reliable source, work through calculation and solution paths yourself, and measure the difficulty against your real class. The time you save goes not into typing but into subject-matter checking — which is exactly where your added value and your responsibility as a teacher lie.

How do I differentiate material with AI?

Differentiation is one of the strongest use cases, because it otherwise costs a lot of time. A language model quickly outputs the same content at several levels: a text in three reading tiers, tasks as "basic", "standard" and "extension", or extra supports for individual tasks. Extension tasks for fast learners are also created in seconds, instead of laboriously varying them by hand.

What matters is to differentiate sensibly — by prior knowledge, reading ability and pace, not by supposed "learner types". The widespread idea of fixed learning styles (visual, auditory, kinaesthetic) is considered unsupported in education research; tailoring material to it gains you nothing in effect. What robust differentiation looks like instead is explored in the piece on differentiated instruction. The AI quickly supplies the variants — which one fits pedagogically is your call.

What about data protection and copyright?

Two sober limits belong here. First: personal data of individual students — names, grades, diagnostic notes — do not belong in a public AI tool. Use anonymised details or systems cleared by your school. Second, AI does not replace clearing rights: if you load in someone else's texts or images, copyright and your school's usage rules remain decisive. When in doubt, follow the guidance of your school or region — the data-protection and AI guidelines are currently being updated on a rolling basis.

Conclusion: AI drafts, you take responsibility

Creating teaching materials with AI is not a replacement for subject-matter work but an acceleration. The model delivers fast rough drafts for worksheets, quiz questions and differentiated versions — checking, selection and pedagogical responsibility stay with you. Anyone who takes this fact-checking seriously gains time without sacrificing quality. You will find more practical help in our For Teachers & Parents category.

And if the checked material should then become something your class uses to practise on its own, scripts and summaries can be turned with tools like LearnCastAI into learning podcasts, flashcards and quiz questions — as an offer for independent practice, not a replacement for your teaching.