Plan from what you already have

Teachers, a randomized trial in England just put a number on AI planning help. Primary teachers using an AI lesson planning tool spent about 2.5 hours a week on lesson and resource preparation, against 3.3 hours for teachers without it. That's roughly 49 minutes a week, and the lesson materials were judged no worse (Education Endowment Foundation).

Here's the part that really stands out to me. The teachers used the tool for only around 7% of their lessons in an average week, and use dropped over the 10 weeks of the trial.

The evaluators offer a reason that makes sense to me. The AI planning tool was built to plan whole lessons from scratch, and in their words, "Teachers did not always need to plan whole lessons from scratch." Most weeks, they were adapting what they already had.

That's the planning most educators do. Most weeks you're taking last year's lesson, or material from your curriculum resources, and making it fit this class after what happened today. New lessons from a blank page are the exception. So this AI lesson planning workflow is built for the adapting.

Set up a planning project once

Most AI tools now have a place to keep files and standing instructions together, so you aren't explaining your class from zero every time. In Claude and ChatGPT it's called a project. Anthropic says projects are available on every Claude plan, including free accounts, which can have up to 5 (Anthropic). OpenAI's help page suggests reusing a project "for recurring tasks" (OpenAI). If your school is on Google, Gemini Notebook does a similar job: you keep your sources in a notebook and Gemini works from them (Google). Use whichever tool your school has approved.

Put these in the project:

• Your unit plan or pacing guide
• The lessons and materials you already have for the unit
• The learning targets or standards for the unit
• A short description of the class: grade, period length, what tends to work, and where this group usually needs more time

Keep names, IEPs, grades and anything about an individual student out of it. Describe the class as a group. If you want more on that boundary, the class data edition walks through it.

Then paste this into the project's instructions box:

Adapt next week's lessons with one short prompt

Each week, open a new chat inside the project and use this:

The first prompt you write once. The second is the weekly habit, and it's short on purpose. Your class context is already saved, so the only new thing you type is what happened this week. That's what makes it easy enough to come back to.

What you should get back

A day-by-day draft built mostly from your own lessons, each piece tagged with the file it came from, a short list of things to make, and questions where it didn't have enough to go on. The file tags are the most useful part. They show you which parts are yours and which parts are new.

How to check an AI-drafted lesson plan

• Hold every learning target against your unit plan. They should match word for word.
• Open every file it says it reused. Make sure the file exists and says what the draft claims.
• Read each check for understanding and ask whether it would tell you who's ready to move on.
• Look at everything marked as added. Keep it, change it, or cut it.

Then borrow one thing from the trial and track your time. Write down how long planning takes this week and for the next 2. If it isn't saving you time by week 3, change the setup or drop it. If a draft keeps missing, how to get better answers from AI covers naming the problem and changing one instruction at a time.

My one warning

Watch the file tags. A plan that says it came from your own materials is the easiest kind to wave through, because it sounds like you planned it. AI tools can describe a file as containing something it doesn't, or blend 2 lessons into one and credit the wrong file. Open the files.

The evaluators said it plainly: time saved doesn't automatically mean better planning, and teachers still need to check and adapt what AI produces. The 49 minutes only counts if the review still happens.

Next week

Feedback comments that point to what the student actually wrote. If you're new here, practical AI for teachers: where to start is the place to begin.

What's the planning job you redo every single week? Reply and tell me. I may build a future edition around it.

Claude is a product of Anthropic, ChatGPT is a product of OpenAI, and Gemini is a product of Google. EdTeacherAI is not affiliated with or endorsed by any of them.