AI Prompts for Teachers: 14 That Save Real Hours

Where the hours actually go

Most AI-for-education content targets the wrong tasks. It offers to write lesson plans from scratch and to grade student work — the two jobs where a teacher's professional judgement matters most and where a generic output is least useful.

The real time sink is different, and less glamorous. It is producing four versions of the same worksheet for four reading levels. It is writing the rubric for an assessment you have already designed in your head. It is turning three bullet points of observation into a parent email that is warm, specific and defensible. It is generating six more practice problems in exactly the format the last six were in.

That work is repetitive, high-volume, and involves almost no judgement once you have made the decisions that matter. It is precisely what AI is good at, and it is where the hours are.

The rule that has to come first: student data

Before any prompt, the boundary. Do not put identifiable student information into a general-purpose AI tool. Not names, not the details that identify a child within a small class, not anything from a special educational needs record, not safeguarding notes, not anything a parent would be alarmed to find outside the school.

This is not caution for its own sake. Consumer AI tools may retain input and may use it for training depending on the tier, and a school's obligations under data protection law do not pause because a tool is convenient. Your institution may also have a policy that is stricter than the law.

The practical workaround costs almost nothing: describe the situation without identifying the pupil. "A Year 8 student who is strong verbally but avoids written work and becomes anxious when asked to share" gets you an equally useful answer to using their name, and takes three extra seconds.

• Never: names, initials with a class, dates of birth, addresses, SEN or medical detail, safeguarding notes, anything from a pupil record.
• Fine: anonymised descriptions of a situation, your own lesson materials, published curriculum documents, work samples with names removed.
• Check your institution's policy before using any tool for anything touching pupils — it may be stricter than you expect.
• If you paste student writing to get feedback ideas, strip the name first and check the text itself does not identify them.

Prompt 1 — Differentiate a task you have already made

This is the highest-value prompt on the page. You have designed one good task; you need three versions of it. The design work is done, and the adaptation is mechanical.

• Prompt: "Below is a task I have written for [YEAR GROUP] on [TOPIC]. Produce three versions: (1) a scaffolded version for students who need more support — same learning objective, more structure, sentence starters where useful, no reduction in cognitive demand; (2) the original, lightly polished; (3) an extension version that increases the demand rather than the volume. Keep all three assessing the same objective. Do not simplify the vocabulary in version 1 unless a word is genuinely a barrier — say which words you changed and why. My task: [PASTE]"

Prompt 2 — Build a rubric from criteria you have decided

Rubrics are slow to write and fast to specify. You already know what you are looking for; the labour is turning that into level descriptors that read consistently.

• Prompt: "Write a four-level rubric for [TASK] with [YEAR GROUP]. The criteria I am assessing are: [LIST YOUR CRITERIA]. Rules: describe what the work shows, not what the student lacks — no descriptor should be phrased as an absence. Each level must be distinguishable from the one below by something observable in the work, not by adverbs like 'mostly' versus 'sometimes'. Write the descriptors in language a student in this year group could read and use. Then list the two places where markers are most likely to disagree."

Prompt 3 — Worked examples at three levels

Worked examples are among the best-evidenced tools in teaching and among the most tedious to produce in quantity.

• Prompt: "Produce three worked examples of [PROBLEM TYPE] for [YEAR GROUP], increasing in difficulty. For each: show every step, and annotate the step where students most commonly go wrong with a one-line note on the misconception behind that error. Use the same notation and layout throughout so students can compare them. After the three examples, give me two problems with no worked solution for students to attempt."

Prompt 4 — Anticipate the misconceptions

Knowing where a class will stumble before the lesson is worth more than any resource. This prompt is planning support rather than content generation.

• Prompt: "I am teaching [CONCEPT] to [YEAR GROUP]. List the six most common misconceptions students hold about this, ordered by how frequently they appear. For each: state the misconception in the words a student would actually use, explain why it is a reasonable thing to believe given what they already know, and give one diagnostic question that would reveal whether a student holds it. Do not give me the correct explanation — I have that. I want the wrong models."

Prompt 5 — Turn observations into a parent email

This is where a lot of evening hours go: converting three honest bullet points into something warm, specific, and safe to send.

• Prompt: "Write a short email to a parent from these notes: [YOUR BULLET POINTS — NO NAMES]. Rules: under 180 words. Open with something specific and genuine, not a generic pleasantry. State the concern plainly in one sentence — do not bury it. Describe behaviour and work, never character. Give one concrete thing the parent could do at home. End with an offer to talk, not a demand for a meeting. Do not use the phrases 'reach out', 'touch base' or 'at this moment in time'."

Prompt 6 — Feedback comments that a student can act on

The point of feedback is the next draft. Most feedback describes the current one, which is why students read the grade and stop.

• Prompt: "Below is a piece of [YEAR GROUP] work on [TASK], with the name removed. Write feedback in three parts: (1) one specific thing that works, quoting the exact phrase and saying why it works; (2) the single most important improvement, phrased as an instruction the student can follow on the next draft; (3) one question that makes them think rather than telling them the answer. Rules: address the work, not the student. No praise sandwiches. Under 120 words. Do not give a grade. Work: [PASTE]"

Prompt 7 — A retrieval-practice quiz that is actually spaced

Quiz generation is easy; quiz generation that pulls from earlier in the term is the part that gets skipped when you are tired.

• Prompt: "Create a ten-question retrieval quiz for [YEAR GROUP]. Six questions on [THIS WEEK'S TOPIC], three on [TOPIC FROM THREE WEEKS AGO], one on [TOPIC FROM LAST TERM]. Mix formats: some recall, some application. For each question give the answer and the one-line explanation a student would need if they got it wrong. Do not make the older questions easier — the point is that they are harder to retrieve, not simpler."

Prompt 8 — Explain a concept five different ways

When an explanation does not land, teachers need a different route, quickly, mid-lesson.

• Prompt: "Explain [CONCEPT] to [YEAR GROUP] in five different ways: (1) a concrete everyday analogy, (2) a visual description I could draw on the board in thirty seconds, (3) a worked numerical or textual example, (4) by contrast with something they already understand, and (5) by starting from a common wrong answer and showing where it breaks. Keep each under 80 words. Tell me which of the five is most likely to introduce a new misconception, and what that misconception would be."

Prompt 9 — Adapt a reading passage without dumbing it down

Reading-level adaptation is one of the most common differentiation tasks and one of the easiest to do badly.

• Prompt: "Rewrite the passage below for [YEAR GROUP] readers who are approximately two years behind. Rules: keep every idea — do not remove content. Shorten sentences and reduce clause density rather than replacing subject vocabulary; technical terms stay and get a brief gloss on first use. Keep the original structure so students can follow along with the unadapted version. List every term you glossed. Passage: [PASTE]"

Prompt 10 — A lesson sequence, not a lesson plan

A single AI lesson plan is usually generic. A sequence with dependencies stated is genuinely useful planning support.

• Prompt: "Outline a sequence of five lessons on [TOPIC] for [YEAR GROUP], working backwards from this end point: [WHAT STUDENTS SHOULD BE ABLE TO DO]. For each lesson give: the specific objective, what students must already be able to do before it, one checkpoint that would tell me mid-lesson whether to move on, and the most likely reason a lesson would need re-teaching. Do not give me starter activities or timings — I will do those."

Prompts 11–14 — The small ones that add up

Not every useful prompt needs a paragraph of setup. These are the short ones worth keeping to hand.

• Report comment bank: "Write eight report comment stems for [SUBJECT] at [YEAR GROUP] covering a range from significant concern to exceptional. Each must describe observable work or behaviour, contain a specific next step, and avoid the words hardworking, lovely, and potential."
• Knowledge organiser: "Produce a one-page knowledge organiser for [TOPIC], [YEAR GROUP]: key vocabulary with student-facing definitions, the five facts they must retain, one diagram described in words, and three questions that test the whole page."
• Practical or activity risk check: "I am planning [ACTIVITY] with [YEAR GROUP]. List what could realistically go wrong, ordered by likelihood rather than severity, with one mitigation each. Include the organisational failures — timing, resources, transitions — not only the safety ones."
• Cover work: "Write cover work for [YEAR GROUP] on [TOPIC] that a non-specialist can supervise: self-contained instructions on the sheet, no equipment, a clear finishing point, and an extension for anyone who finishes early. Include a two-line note to the cover teacher explaining what students should already know."

What to keep away from AI

Being clear about this is what makes the rest usable in a professional setting.

• Grading decisions. Use AI to draft feedback wording from your judgement; do not let it form the judgement. A model has not taught this child and cannot weigh what you know about them.
• Anything identifying a pupil. Covered above, and worth repeating because it is the mistake most easily made in a hurry.
• Safeguarding. These are institutional processes with defined routes. An AI tool is not part of that route in any circumstance.
• Facts you will teach without checking. Models fabricate confidently, and a plausible wrong date or misattributed quotation delivered in a lesson is very hard to unpick afterwards.
• References and sources. The most reliably fabricated category. If a prompt returns a citation, open it before it reaches a student.
• Anything that removes your professional voice from a communication a parent will read as coming from you personally.

Prompt 15 — The admin nobody trained you for

A meaningful share of a teacher's week is documentation that has nothing to do with teaching: the intervention record, the trip risk assessment, the departmental report, the evidence log for an inspection, the funding application nobody has time to write. It is unglamorous, it is compulsory, and it is where the hours vanish without anything to show for them.

This is the safest category of AI use in a school, because it usually involves no student data and the output is read by adults who will edit it. It is also the category teachers think of last.

• Prompt: "Turn these notes into a [DOCUMENT TYPE] for [AUDIENCE]. Notes: [PASTE]. Requirements: use the headings [LIST]; keep it under [N] words; write in plain professional English with no educational jargon that a governor would not recognise; where a section needs information I have not given you, write [MISSING: what is needed] rather than inventing plausible content. Do not add recommendations I did not make."
• For repeated documents, ask once for a reusable skeleton: "Produce a blank template for this document with a one-line prompt under each heading telling me what to write there." Fill the template in future terms instead of regenerating the whole thing.

Teaching students to prompt, instead of policing them

The enforcement approach to AI in classrooms is losing, and detection tools have a false-positive problem serious enough that acting on them can be unjust to the student who did nothing wrong. The more durable response is to move the assessed thing to somewhere AI cannot reach, and to teach the skill openly.

Assess the process rather than the artefact. Ask for the drafts, the prompts used, the critique of what came back and the reasoning behind what was kept. A student who submits a transcript of their prompting alongside their final piece has demonstrated more thinking than one who wrote the same piece unaided, and you can see it.

The prompts below are ones to hand to students rather than to run yourself. Each turns the model into something that will not do the work for them.

• The Socratic tutor: "You are helping me understand [TOPIC]. Do not give me answers. Ask me one question at a time to find out what I already understand, then ask the next question that moves me forward. If I am wrong, do not correct me — ask a question that helps me notice."
• The adversary: "Here is my argument: [PASTE]. Give me the three strongest objections a well-informed person would raise. Do not rewrite my work and do not tell me it is good."
• The explainer check: "I am going to explain [CONCEPT] to you. Point out only what is inaccurate or missing. Do not add your own explanation." This is retrieval practice with instant feedback, and it is the single most useful classroom prompt there is.
• The source auditor: "Here is a claim an AI made: [PASTE]. What would I need to check to know whether it is true, and where would I look?"

A departmental prompt bank beats fifteen private ones

The pattern in schools that get real value from this is not that individual teachers become skilled prompters. It is that one person writes a good prompt, it goes into a shared document with the subject and year group in the title, and fourteen colleagues stop reinventing it.

The barrier is almost never technical. It is that a prompt sitting in someone's chat history is invisible, and nobody shares something they think of as a throwaway message. Treating prompts as departmental resources — the same status as a scheme of work or a shared assessment — changes that in a term.

Keep the bank small and curated rather than exhaustive. Twenty prompts people actually use beats two hundred nobody can find, and a prompt that gets edited every time it is used should be fixed in the bank rather than patched by each person separately.

• Store each prompt with three things: the prompt itself, one example of a good output, and a one-line note on what to check before using the result.
• Name them by job, not by technique: "Y9 differentiate a worksheet" is findable, "few-shot differentiation prompt" is not.
• Bake your school's constraints in once — the exam board, the marking policy, the reading age targets, the banned phrases — so nobody has to remember them.
• Review the bank once a term. Delete what nobody used, and promote whatever people have been quietly pasting from a colleague's email.
• Agree the data rule in writing at the top of the document, so the answer to "can I paste this?" is one scroll away rather than a conversation.

Making these repeatable

The prompts above are worth more the second time than the first, and that only happens if you stop rewriting them.

Replace the specifics with fields — [YEAR GROUP], [TOPIC], [CONCEPT] — and save them somewhere you can reach in the ninety seconds you have between lessons. A prompt you have to reconstruct from memory is a prompt you will not use on the day you most need it.

The differentiation and rubric prompts also chain naturally: build the rubric first, then feed it into the feedback prompt so the comments reference the same criteria the students were assessed against. That consistency is hard to maintain by hand across thirty pieces of work and trivial to maintain when the rubric is the input to the next step.

One last suggestion: keep your own exclusion list. Every time an output uses a phrase that would make you wince in front of a class or a parent, add it to a standing "do not use these words" line. Within a term that list is doing more for the quality of the output than any other part of the prompt.

Related articles