How AI Turns One Standard Into a Week of Lessons You Can Actually Run
Some links below are affiliate links — I may earn a small commission, at no extra cost to you, if you sign up for a tool I recommend. More on how I choose what to recommend.
It’s the Thursday of the week before. The standard is open in one tab, an empty planning template in another, and you have five periods to fill. You know roughly where the unit has to end. You have no idea yet what Tuesday looks like.
So you do what everybody does. You open ChatGPT and type: “Write me a week of lesson plans on cell division for 7th grade.”
What comes back looks like a lesson plan the way a stock photo looks like your classroom. Day 1: introduce the topic, activate prior knowledge. Day 2: engage students in discussion. Day 3: small groups. Day 4: review. Day 5: assess. None of it is something you can actually do at 8:42 on Tuesday.
That isn’t AI being bad at lesson planning. It’s AI doing exactly what you asked. You gave it a topic. What it needed was constraints.
Why “plan me a week” always fails
A topic has no shape. “Cell division for 7th grade” could be a lecture, a lab, a card sort, a semester. With nothing to aim at and nothing to work within, the model produces the average of every lesson plan ever written — definitionally, the generic one.
Two things fix that, and they compound.
Half one: work backward from the assessment. Don’t give AI the topic. Tell it what a student has to be able to do on Friday, and ask it to plan the days that get them there. Backward design isn’t new; what’s new is that you can hand the tedious middle of it to a machine.
A plan built forward from a topic wanders — it covers things. A plan built backward from a demonstrable skill has a spine, and every day has an obvious job: Monday builds the vocabulary Friday’s task requires, Tuesday practices the first half of the move, Wednesday the second, Thursday puts them together under low stakes.
Half two: feed it your actual room. This is the biggest difference between output a teacher deletes and output a teacher runs, and almost nobody does it.
Your room has facts. Periods are 47 minutes, not “a class period.” The laptop cart is yours on Wednesday only. There’s no printing budget past October. Your fourth period will not sit through twenty minutes of lecture. Six students have read-aloud accommodations. There’s a fire drill Thursday.
Every one of those facts you withhold is a place the plan will be useless. AI can’t guess them. Tell it, and it plans around them.
The four things AI is genuinely good at here
Planning is professional judgement. These four parts are typing.
Sequencing days backward from a target skill
Give it the Friday task and it works out what has to be true by Thursday, and by Wednesday for that. It's good at this because it's a logic problem, not a creative one — and it's the part that costs you forty minutes with a whiteboard.
Fitting activities to constraints you name
Tell it you have 47 minutes, no laptops until Wednesday and a class that needs to move every twelve minutes, and it stops suggesting the 30-minute video. Constraints don't limit the output — they're what make it specific.
Naming the evidence for each day
Ask what you should see at the end of Tuesday and you get an exit-ticket-level answer, not a learning objective. "Three of four students label the phases in order without the diagram" is checkable. "Students will understand mitosis" is not.
Writing the contingency nobody writes
"If half the class doesn't have it, do this instead tomorrow." Published curricula rarely include this and generic AI output never volunteers it — but it's the part you reach for, because the day half the class doesn't have it is most days.
The before and after
The vague version:
Write me a week of lesson plans on cell division for 7th grade science.
Five days of headings. A simulation you don’t have licences for, a poster project that eats two days and produces nothing assessable, and a Friday test unconnected to anything Monday through Thursday did.
The constrained version is the prompt in the next section. What changes:
- Every day is timed to 47 minutes — opener, main block, close — and the minutes add up.
- Nothing needs the laptop cart except Wednesday, because you said so.
- Thursday is deliberately short, because you mentioned the drill.
- Monday’s vocabulary is exactly what Friday’s task requires. Nothing extra, nothing missing.
- Each day ends with a named piece of evidence and an “if they don’t have it” branch.
- Wednesday is flagged as likely to run long and Tuesday short, so you know where the slack is before you’re standing there with nine minutes left.
Same model, same day, thirty seconds of extra typing.
The exact prompts to use
Copy these into ChatGPT, Claude or Gemini. The free tier handles this.
1. The backward-design week
Fill in every bracket. The brackets are the point — an empty one is a place the plan drifts.
I teach [grade] [subject]. Help me plan one instructional cycle using backward design.
The standard, in full text (not the code): [paste the sentence from your framework]
What students must be able to DO at the end: [Friday’s task, concretely — “write a paragraph arguing a claim using two pieces of text evidence,” not “understand argumentative writing”]
My actual room:
- [N] periods this cycle, [N] minutes each; class size [N]
- [N] students with accommodations: [read-aloud / extended time / etc.]
- [N] students reading noticeably below grade level
- Technology I actually have: [e.g. shared laptop cart, Wednesdays only; projector daily]
- Materials constraints: [e.g. no printing budget, no lab space]
- What this class won’t tolerate: [e.g. more than 10 minutes of me talking]
- Known disruptions: [fire drill Thursday; assembly shortens Friday by 15 min]
- What the previous unit covered: […]
What I want back:
- A table, one row per day: Day | Objective (one student-facing sentence) | Opener (minutes) | Main activity (minutes) | Close (minutes) | Materials to have ready | Evidence I should see by the end of the period.
- Each day’s minutes must add up to my period length. Do not exceed it.
- One sentence per day explaining how it feeds the Friday task. If a day doesn’t clearly feed it, cut the day and tell me.
- Flag which day is most likely to run SHORT and which to run LONG, with one extension for the short day and one thing to cut on the long day.
- Use only the materials and technology listed above. Nothing else.
- At the end, tell me what you were unsure about.
2. Expand one day into a run sheet
Once the week holds up, take the day you’re least sure about and get it minute by minute — the version you could teach half-awake.
Take Day [N] from the plan above and expand it into a minute-by-minute run sheet for a [N]-minute period.
Include: the exact words I say to launch each part (2–3 sentences each), what’s on the board when students walk in, what I’m doing while they work, the questions I circulate and ask, transition cues, and what goes home.
Mark anything I must prep before the bell with a PREP tag and list all PREP items at the top.
Finally: where am I most likely to lose the room, and what do I do there?
3. The contingency pass
Short, and the one teachers keep.
For each day of the plan: if I check the exit evidence and roughly half the class doesn’t have it, what do I do at the start of the next period — in 10 minutes or less — without abandoning the rest of the week? Be specific about what I re-teach and what I let go.
Also, which single day, if it goes badly, most endangers the Friday assessment? That’s the day I’ll build the most slack into.
A realistic 25-minute workflow
Write Friday first, then the constraints (9 min)
Before you open anything, describe the end-of-cycle task in one concrete sentence: what students produce, under what conditions. If you can't write it, the week isn't ready to plan. Then fill prompt 1 in, pettily — period length to the minute, which days you have the cart.
Generate, then read the table by column, not by day (5 min)
Read the Evidence column top to bottom first. If those five items don't build toward Friday's task, the sequence is wrong and no activity-polishing fixes it. Then read the Materials column and cross off anything you don't have.
Cut something (3 min)
It has over-planned. It always over-plans. Assume every main activity runs a third longer than stated and trim — usually by deleting the second half of the most ambitious day.
Run the contingency prompt (3 min)
Prompt 3. Paste the answers into your planner's notes column. You will use at least one this week.
Expand the day you're least sure about (5 min)
Prompt 2, on one day only. You don't need a script for the day you've taught eleven times — you need one for the day with the new activity in it.
Layer the differentiation on afterwards
The plan tells you what happens. It doesn’t tell you how the kid reading three years below grade level gets into Tuesday’s text. Do that as a second pass — asking for both at once produces a plan that’s vague about both.
Once the sequence is settled, run each day’s reading through turning one worksheet into five reading levels: same objective, different points of entry, vocabulary locked. And if Friday is a quiz, four versions of it take about a minute.
What AI can’t do (the limits to know)
A few common mistakes the prompts will catch
Planning on a schedule where this falls apart — block periods, co-taught sections, a rotating timetable? Email help@aiforyourday.com. I read every one, and the cases teachers send are what make the next version of this article better.