AI prompts for teachers — write a prompt that works
AI tools (ChatGPT, Claude, LectaMe) are only as good as the prompt you give them. A vague prompt yields vague output; a targeted prompt yields directly usable lesson content. Below are five concrete prompt patterns for lesson preparation, cases, quiz questions, feedback and differentiation — each with a “bad versus better” example so you can see the difference.
The four elements of a good prompt
- Context: level, subject, audience, lesson type. (“For HAVO 4 biology, a 50-minute lecture”)
- Structure: what format should the output have? (“Constructive alignment principle, 3 learning objectives, theory + case + check”)
- Tone: how should it sound? (“Concrete, no nonsensical business jargon”)
- Constraints: what should and shouldn't be included? (“Max 100 words, no smileys, friendly tone”)
Five prompt patterns for teachers
Create a lesson plan
BAD
"Create a lesson about photosynthesis"
BETTER
"Create a 50-minute lesson plan about photosynthesis for HAVO 4 biology. Follow the constructive alignment principle: start with 3 SMART learning objectives, then a 15-minute theory block with a visual explanation of the Calvin cycle, a case with a real-world example (why plants turn yellow in autumn), 3 formative check questions, and close with an exit ticket. Use a concrete language level suitable for HAVO."
Why: The good prompt provides: level (HAVO 4), duration (50 min), didactic framework (constructive alignment), structure (theory + case + check + exit), a concrete example, and a language-level instruction.
Generate a case
BAD
"Give me a nursing case"
BETTER
"Write a patient case for MBO nursing level 4, for the lesson on clinical reasoning in dyspnoea. Build it up using the SBAR format. Patient: woman, 78 years old; start with a general presentation, then vital parameters (tachypnoea, mild hypoxia, raised heart rate), history elements that lead to three possible diagnoses (cardiac decompensation, pneumonia, pulmonary embolism). End with an open question for the student: what additional information do you need? Do not give the diagnosis."
Why: A specific level, lesson topic, format (SBAR), building principle (three differential diagnoses), and a didactic instruction (open ending, no solution).
Create quiz questions
BAD
"Give me 5 quiz questions about algebra"
BETTER
"Write 5 multiple-choice questions for VMBO 3 mathematics on solving linear equations. Question levels: 2 questions on facts/recognition, 2 on application (solving a calculation), 1 transfer (word problem). Per question, 4 answer options of which 1 is correct and 3 are plausible distractors (no nonsensical options). For each question add: the learning objective it tests (e.g. "can solve an equation with x") and an explanation of why the correct answer is correct."
Why: Level, topic, question distribution (facts/application/transfer = Bloom's taxonomy), a quality rule for distractors, and a link to the learning objective.
Formulate feedback
BAD
"Give feedback on this student response"
BETTER
"Give feedback on this student response (HAVO 4 history, assignment on the Industrial Revolution). Response: [text]. Assess: (1) historically correct, (2) use of sources, (3) structure of the argument. Use a feedback sandwich (strong point → point for improvement → motivating close). Maximum 100 words. Write in a friendly but substantive tone — not childish, no smileys. Steer towards what the student can concretely do differently, not a generic "think harder"."
Why: Context (subject, level, assignment), assessment criteria, format (sandwich), length limit, tone instruction, and a specific direction for improvement.
Devise a differentiation layer
BAD
"Make this lesson for different levels"
BETTER
"Here is my lesson content on the Pythagorean theorem for VMBO 3 [text]. Build two scaffolding levels: (1) BASIC layer for students who struggle — split the explanation into 3 micro-steps with concrete examples from everyday life (a staircase, a roof angle). (2) ADVANCED layer for stronger students — give a challenging application (distance between 2 points on a coordinate system) plus a short historical context (why it is named after Pythagoras). Keep the basic layer concise, no unnecessary words."
Why: Two scaffolding directions made explicit, with an instruction on how they differ in depth, and a didactic model (basic vs advanced) applied consistently.
Frequently asked questions
Which AI tool is best to use as a teacher?+
Three main categories. (1) ChatGPT/Claude for brainstorming, writing and general didactic questions. (2) LectaMe for building complete interactive presentations with a didactic structure (theory/case/quiz). (3) Specific tools for images (LectaMe Image Studio, Midjourney) or video (LectaMe Videostudio, Descript). Most teachers benefit from 1-2 tools. Also read /voor-docenten/ai-tools for combinations.
How do I know whether an AI answer is factually correct?+
Three checks. (1) For general concepts: AI is usually accurate, but when in doubt always check a reliable source (a textbook, NRO, scientific literature). (2) For specific facts (dates, names, quotes, numbers): AI often hallucinates here — always verify. (3) For subject matter you know well: trust your instinct. AI is a faster typist, not a peer expert. For MBO/HBO clinical context: compare with verified medical sources.
Am I allowed to use AI-generated lesson content freely?+
For educational use in your own lessons: almost always yes. But note: (1) Factual accuracy remains your responsibility. (2) For commercial use (a book, a paid course) check the licence of the specific AI tool. (3) AI may not copy the creative property of a specific author — always use output as a basis, not as finished work. Read /ai-belofte for LectaMe-specific rules.
How long should a good prompt be?+
50-200 words for most lesson-preparation questions. Shorter gives too little context (AI guesses too much); longer overwhelms the model and paradoxically produces worse output. Rule of thumb: one prompt = one clear instruction + context + an example of what you want. For complex work: split it into several prompts ("first this, then that") instead of one long one.
What do I do if the AI doesn't do what I want?+
Three strategies. (1) Give an example of GOOD output that you do want to see (few-shot prompting). (2) Negative instructions work — "no smileys", "no summarising closing sentence". (3) Set a role — "You are an experienced MBO nursing teacher". (4) Refine iteratively: use the first answer as input for a follow-up prompt: "Make this shorter and more concrete".
How do I prevent AI from giving boring or generic output?+
Four tips. (1) Give context about your audience (level, subject, class energy). (2) Ask for concrete examples from students' everyday lives. (3) Set a tone direction ("lively", "witty", "challenging") — avoid "professional" because that triggers generic business language. (4) Use AI output as a starting point, not as a finished product: apply your own voice and examples.
Sources
- Kennisnet — Dutch authority on ICT in education, including practical AI tips.
- Vernieuwenderwijs.nl — blogs on applying AI in Dutch educational practice.
- Learn Prompting — open-source community and course on prompt engineering.