ChatGPT Prompts for Teachers: Lesson Plan Prompts That Survive a Real Classroom

The short version: the ChatGPT prompts that save teachers real time aren't the generic "write me a lesson plan on photosynthesis" type, they're the ones that ask for structure, timing and differentiation in the same breath. I spent six weeks testing prompts with a secondary school teacher and got her Sunday night planning down from three and a half hours to under an hour. Below are the exact prompts, the process, and the bit nobody selling AI to schools wants to say out loud.

Why the average "20 prompts for teachers" listicle falls apart in week two

I've read most of the popular teacher prompt lists floating around LinkedIn and Pinterest. Nearly all of them read like they were written by someone who has never stood in front of thirty Year 9s on a wet Friday afternoon. The prompts ask ChatGPT to "create an engaging lesson plan" and stop there. What comes back is a lesson that would work in a curriculum textbook and falls apart the moment a real class doesn't behave like the fictional one in the AI's imagination.

A lesson plan that survives a real classroom needs timing built in, a plan for the kid who finishes in four minutes, a plan for the kid who needs the whole hour, and a way to recover if the starter activity runs long. Most listicles skip all of that because it's harder to write and less satisfying to skim.

That's the gap I try to close. The prompts below aren't shorter or cleverer than what's already out there, they're just built around what a lesson plan has to do once it leaves the screen.

The prompts I give teachers, organised by the job, not the subject

Subject-specific prompts are a waste of time because you'll only reuse them once. These are organised by what stage of planning you're at, which means you can reuse the same fifteen prompts for every subject and every year group you teach.

Scheme of work and long-term planning

"I teach [subject] to [year group/age] in the UK. Build me a six-week scheme of work on [topic], broken into individual lessons, each with a learning objective, a suggested duration, and one formative assessment point. Assume 50-minute lessons and a class of mixed ability with at least three pupils on SEN support plans."

"Take this scheme of work [paste it] and tell me where the pacing is unrealistic for a 50-minute lesson. Flag any lesson that's trying to cover too much."

"Turn this six-week scheme into a one-page overview I can print and stick on the classroom wall, with just the topic, key question and homework for each week."

Single lesson prompts

"Write a 50-minute lesson plan on [topic] for [year group]. Include a five-minute retrieval starter linked to last lesson, a main activity with a clear time allocation, a mini-plenary at the midpoint, and an exit ticket with three questions. Write it as a table I can print."

"I have 15 minutes left in this lesson and the planned activity has overrun. Give me two quick, low-prep options to close the lesson that still check understanding of [topic]."

"Rewrite this lesson plan assuming the projector and internet are down. No videos, no slides, no online quiz tool."

Differentiation prompts

"Here's my lesson plan [paste it]. Give me three versions of the main task: one scaffolded for a pupil reading two years below age-related expectations, one for the middle of the class, and one extension task for a pupil who finishes early and needs stretch, not just more of the same."

"Suggest sentence starters and a word bank for [topic] suitable for an EAL pupil at an early intermediate level."

"Take this worksheet and rewrite the instructions in plain, short sentences without changing the content, suitable for a pupil with dyslexia."

Starters, plenaries and retrieval practice

"Give me five retrieval starter questions on the last three topics I've taught this half term in [subject], each answerable in under two minutes."

"Write a low-stakes multiple choice quiz, eight questions, on [topic], with the correct answer and one common misconception built into the wrong answers."

"Suggest three plenary activities that take under five minutes and don't need any printed resources."

Marking and feedback prompts

"Here is a mark scheme for a GCSE-style question [paste it]. Write three model answers, one grade 4, one grade 6, one grade 8, so I can show pupils the difference in quality."

"Draft feedback comments for a piece of work that has good content but weak structure. Keep it specific and actionable, no generic praise."

If you want a broader set of these, I put together a longer, subject-neutral list in my post on ChatGPT prompt limits and how to get more out of each one, which is worth reading before you start typing prompts into the free version and hitting the daily cap halfway through your planning session.

A term of Year 9 planned at a kitchen table

Last spring a friend of mine, a history teacher in Essex called Ruth (name changed, the school didn't sanction this so I'm keeping her out of it), asked me to help her get ahead of a Year 9 unit on the Cold War before half term. She was doing her usual Sunday routine: laptop open, tea going cold, four hours minimum, and she still felt behind by Monday.

We sat down for ninety minutes on a Saturday. I didn't teach her anything about history, I don't know enough to. What I did was build her a chain of prompts: scheme of work first, then each individual lesson pulled from that scheme one at a time, then a differentiation pass on the two lessons she knew would be hardest for her lower set. We fed her existing resources in as context rather than starting from nothing, which mattered more than any clever prompt wording.

By the end of that session she had six weeks of lessons, each with a starter, main activity, and exit ticket, plus three differentiated versions of the two trickiest lessons. She still spent time on it every week after that, because no plan survives contact with a real class untouched, but her Sunday sessions dropped to around 45 minutes of adjusting rather than three and a half hours of building from a blank page. That's not a hypothetical saving, that's what she told me six weeks later when I checked in.

The part worth saying plainly: the time saved wasn't in the writing. ChatGPT writes fast regardless. The time saved was in not staring at a blank document trying to remember the structure of a good starter activity at 9pm on a Sunday. That's the real cost AI removes, decision fatigue, not typing speed.

For school holidays, I've written ChatGPT prompts for parents to keep kids busy.

The six-step process for planning a week in one sitting

1. Paste your scheme of work or curriculum outline into ChatGPT as context before you ask for anything, even if it's messy or in bullet points. A prompt with no context produces a generic lesson every time.
2. Ask for one lesson at a time, not five at once. Batch requests produce lessons that all feel the same because the model reuses the same structure to save effort.
3. Specify your exact lesson length and class size in every prompt. "50 minutes, class of 28" produces a different, more usable answer than no constraint at all.
4. Ask for the output as a table, not prose. Tables are faster to scan on a Monday morning and easier to print.
5. Run a differentiation pass separately, after the base lesson is done, rather than asking for everything in one giant prompt. One clear question gets a better answer than five stacked into a paragraph.
6. Read every single output before it goes anywhere near a class. This isn't optional and it isn't about trusting the tool less, it's about the fact that AI has no idea which of your pupils struggles with reading fluency or which two boys can't sit near each other.

That whole sequence, for five lessons, takes around 40 to 60 minutes once you've done it a few times and have your own saved prompt templates. The first time you try it, budget ninety minutes and don't expect to be fast.

What ChatGPT will not do for you, and why that's the point

Here's the part most posts on this topic skip past. AI does not know your class. It doesn't know that Jayden sits by the window and needs to be moved before the main activity starts, it doesn't know the school's behaviour policy, and it has no idea that last week's fire drill ate fifteen minutes and you're now a lesson behind. Every plan it gives you is written for an average classroom that doesn't exist.

The uncomfortable bit is this: a lot of teachers are using ChatGPT to skip the thinking they'd normally do about their specific pupils, not just the typing. A lesson plan that's beautifully structured but ignores the fact that half the class failed the last assessment on this exact topic isn't saving anyone time, it's just producing a nicer-looking version of the wrong lesson. The saving only shows up when a teacher who knows their class uses AI to handle the structure and formatting, and still does the thinking about who's in the room.

I'd also say this plainly because nobody selling AI training to schools tends to: differentiation is the one place AI helps less than people expect. It can produce a scaffolded worksheet in seconds, but it can't tell you if that scaffold is pitched right for the specific pupil you have in mind, because it's never met them. You still have to check it, tweak it, and sometimes bin it entirely. That part of the job doesn't shrink much, no matter how good the model gets.

Getting your school onside before you start

Most UK schools still don't have a written AI policy, which means teachers are quietly making their own rules. That's risky, not because ChatGPT is dangerous, but because if a parent complains about how a piece of homework feedback was generated, "I didn't know we weren't allowed" is a bad position to be in. Check your school's data protection guidance before pasting anything with pupil names or identifiable work into any AI tool, and keep your prompts to generic topics and anonymised examples until there's a clear policy in writing.

If your school is serious about rolling this out across a department or a whole staff, it's worth treating it as a project rather than something each teacher figures out alone on a Sunday night. I map out exactly this kind of rollout for organisations through structured AI implementation coaching, and the schools that do best are the ones that train everyone at once rather than letting adoption happen department by department, six months apart, with wildly different levels of skill.

Teachers picking this up on their own tend to underuse ChatGPT because they treat it like a search engine rather than something you can have a back-and-forth conversation with. If you've ever used AI to learn something yourself, the pattern is the same one I cover in my piece on using ChatGP

• OpenAI's own platform documentation