How to Use AI to Create Lesson Plans Step by Step: A Teacher's Tested Guide

There is a specific kind of Sunday afternoon that every teacher knows.

You have lesson plans due for the week. You have a rough idea of what needs to happen in each class. You open a blank document and then sit there for twenty minutes doing nothing useful while the afternoon disappears. Not because you do not know your subject. Not because you do not care. But because translating what you know needs to happen into a structured, complete, usable lesson plan is a different cognitive task from teaching itself and one that nobody really trains you for.

I spent eleven weeks building and refining a step by step process for using AI to create lesson plans that are actually usable. Not the generic output that every AI produces when you type "lesson plan on fractions." Not a five activity list with no timing and no learning objective worth looking at. A complete, classroom ready plan that accounts for your students, your constraints, and your standards.

This guide is that process. Six steps, in order, with the exact prompts that work and the exact places where most teachers go wrong.

What Makes an AI Lesson Plan Actually Usable

Before the steps, the standard. Because "AI lesson plan" means different things to different tools and most of them are too low.

A usable lesson plan has a measurable learning objective tied to a specific standard. It has an opening that activates prior knowledge or creates genuine curiosity. It has an instructional sequence with realistic timing. It has a formative assessment checkpoint that tells you whether students understood the lesson's central concept before they leave. And it has differentiation built into the activities themselves, not tacked on as a note at the bottom.

That is the benchmark. Every step in this guide is designed to produce output that meets it.

The Data and Privacy Check Before Anything Else

Student information does not go into AI tools unless those tools are covered by your school or district data processing agreement. This applies to lesson planning too. If you are planning for a specific student with an IEP, a named ELL student, or any identifiable individual, describe them in general terms in your prompt. Generic descriptions produce better prompts anyway. Save names and specifics for your own planning documents.

This takes no extra time. It is just the right way to work.

Step One: Write Down What You Actually Know Before You Open Any Tool

This step is the one most teachers skip and the reason most AI lesson plans feel generic.

Before you open Claude or MagicSchool or any other platform, spend five minutes writing answers to these four questions in a notebook or a plain document:

What do my students already know coming into this lesson?

What specific misconception or gap am I trying to address?

What should a student be able to do or explain at the end that they cannot do now?

What constraints does this lesson have to work within, including time, materials, technology access, and any student needs I am planning around?

These four answers are what separate a specific, usable AI lesson plan from the generic template output that wastes everyone's time. The AI cannot know your students. You do. These answers are how you put that knowledge into the process before generation begins.

Time for this step: five minutes. Non negotiable regardless of time pressure.

Step Two: Choose the Right Tool for What You Need

Different AI tools produce different kinds of lesson plans and knowing which to reach for saves significant time.

MagicSchool AI is the fastest option for a structurally complete lesson plan. You fill in a form, specify the grade level, subject, standard, and a few constraints, and it produces a plan with a warm up, instructional sequence, formative check, and closure in about ninety seconds. The differentiation features are built in. The standards alignment is automatic. For most lessons on most weeks, this is the right starting point.

Claude produces the most instructionally sophisticated lesson plans when you give it the right prompt. Where MagicSchool generates reliably and quickly, Claude generates more creatively and with more genuine instructional thinking when you describe what you need in specific terms. For the lessons where you want something genuinely fresh, a unit launch, a concept students historically struggle with, or a lesson that needs a particularly strong hook, Claude is worth the additional prompting time.

ChatGPT on the free tier works with careful prompting but defaults to generic output more readily than either of the above. Use it if it is what you have. Prompt more specifically than you think you need to.

The choice does not matter as much as the prompt. A specific prompt in any tool beats a vague prompt in the best tool every time.

Step Three: Write the Prompt Using Your Step One Notes

This is where the lesson plan is actually built. The prompt is everything.

Here is the full template that produced the strongest lesson plans across eleven weeks of testing:

"I am a teacher planning a lesson. Here is my context. Grade level: [your grade]. Subject: [your subject]. Standard being addressed: [paste the full standard text, not just the code]. What students already know: [your Step One answer]. The misconception or gap I am addressing: [your Step One answer]. What I want students to be able to do by the end: [your Step One answer]. Constraints: [time available, materials, technology access, any student population considerations from Step One]. Now design a complete lesson plan with a warm up activity that activates prior knowledge without giving away the lesson's central concept, an instructional sequence with realistic timing for each segment, a student facing task that requires genuine thinking rather than recall, a formative assessment checkpoint that tells me whether students understood the lesson's main idea before they leave, a closure that connects today's learning to what comes next, and differentiation built into the activities themselves for two levels: students who need additional support and students who are ready for extension. Explain the instructional reasoning behind your activity choices in one sentence per activity."

That last instruction, explain the reasoning, is the single most valuable addition to any lesson plan prompt. It forces the AI to make its pedagogical thinking visible, which means you can evaluate whether the thinking is sound before you use the plan. If the reasoning does not make sense to you as an experienced teacher, the activity probably needs to change.

Time for this step: five to eight minutes to write the prompt, thirty seconds for generation.

Step Four: Run the Checklist Before You Accept the Output

Do not move this output into your planning documents until you have run these checks. They take four minutes and they catch the problems that end up as bad lessons.

Learning objective check: Is the objective measurable? Can you describe what a student would do or produce that would demonstrate they have met it? "Students will understand photosynthesis" is not measurable. "Students will be able to explain the role of chlorophyll in converting light energy to chemical energy using evidence from a data set" is.

Standard alignment check: Does the lesson actually address the standard it claims to address, or does it just mention the standard in the objective and then teach something adjacent? Read the standard again after reading the plan. They should match.

Timing check: Add up the times assigned to each activity. Do they fit your actual period? AI tools frequently generate lessons that run five to ten minutes over. Catch this before class, not during it.

Formative assessment check: Is the formative checkpoint a genuine assessment of understanding or is it a participation activity? A thumbs up or down is not a formative assessment. An exit question that requires students to apply the concept in a new context is.

Differentiation check: Is the support for below level students actually supportive rather than just simpler? Is the extension for above level students a genuine intellectual challenge rather than more of the same work? Generic differentiation notes at the bottom of a plan are not the same as differentiation built into the activity.

Real classroom check: Could you hand this plan to a substitute teacher you have never met and would they know what to do? If the answer is no, the plan needs more specificity before it is done.

Time for this step: four minutes per plan.

Step Five: Add What Only You Can Add

This step takes five minutes and transforms a technically complete lesson plan into one that is actually yours.

AI does not know your students. It does not know that three of them had a heated argument about the ethical implications of the topic last Tuesday that you could build on. It does not know that the kinesthetic activity it suggested will not work with your classroom layout. It does not know that the example it chose is culturally distant from your students' experience and that you have a better one.

After running the checklist, read the plan once more with your students specifically in mind. Add, remove, or adjust anything that the AI generated without knowing what you know.

In my own testing this step produced an average of two to four changes per plan. None of them were large. All of them made the plan meaningfully better for the actual students in the room.

A lesson plan AI generates for a fictional classroom. You teach a real one. This step is where those two things meet.

Time for this step: five minutes.

Step Six: Save the Prompt, Not Just the Plan

This is the step that pays forward to every future planning session and that almost nobody does.

When a prompt produces a strong lesson plan, save the prompt itself in a dedicated document. Label it by subject, grade level, and lesson type. Build a prompt library.

Within four to six weeks of consistent use, you will have a collection of tested, refined prompts that produce strong output for your most common lesson types. The per lesson planning time drops further because the prompt writing time from Step Three becomes largely a matter of updating your existing prompt rather than building from scratch.

At the end of my eleven week testing period I had twenty three saved prompts covering my most common lesson types across three subjects. A week of planning that had taken five to eight hours before the workflow took two hours and ten minutes in the final week of testing. The prompt library was the biggest single factor in that reduction.

Time for this step: two minutes to save. Pays forward indefinitely.

The Full Workflow at a Glance

Step One: Write down what you know before opening any tool. Five minutes.

Step Two: Choose MagicSchool AI for speed, Claude for instructional complexity. One minute.

Step Three: Write the full context prompt using your Step One notes. Five to eight minutes.

Step Four: Run the six point checklist on the output. Four minutes.

Step Five: Add what only you can add. Five minutes.

Step Six: Save the prompt for future use. Two minutes.

Total per lesson: eighteen to twenty five minutes for the planning and review work, not including the reading and preparation time that exists regardless of AI use.

Total per week for five lessons: roughly two hours. Compared to the five to seven hours a well planned teaching week previously cost most teachers I have spoken with.

The Prompts by Lesson Type

These are the specific prompt adaptations that worked consistently across subjects in testing. Use these as starting points and modify for your context.

For a concept introduction lesson, add to the template: "This is the first time students are encountering this concept. The hook should create a genuine need to know without revealing the concept's name. Prioritise curiosity over coverage in the opening fifteen minutes."

For a skill practice lesson, add: "Students have been introduced to this skill but are not yet fluent. The lesson should provide structured practice that gradually releases responsibility from the teacher to the student across the period. Include a specific moment where students check their own work against a clear success criterion."

For a review and consolidation lesson, add: "Students have covered this material but misconceptions remain. Design activities that surface and address those misconceptions directly rather than re teaching the content from the beginning. Students should do the thinking, not receive the summary."

For a project launch lesson, add: "This lesson launches a multi week project. By the end, every student should understand what they are making, why it matters, and what their first concrete action step is. Include time for genuine student questions and build in a hook that makes the project feel worth doing."

What This Workflow Does Not Solve

Honest accounting.

It does not solve the preparation time that exists before lesson planning begins. Reading student work, reviewing assessments, understanding where your class actually is relative to where you need them to be, these are not planning tasks. They are teaching tasks. AI does not reduce them.

It does not replace subject knowledge. A lesson plan generated without deep content knowledge produces plans that look right and teach wrong. The checklist in Step Four catches some of this. Your own subject expertise catches the rest. If you are teaching outside your specialism, apply Step Five with extra care.

It does not produce perfect plans. Every AI generated lesson plan needs human review. The checklist is the minimum. Your professional judgment is the standard.

And it does not solve the problem of a lesson plan that was right for the plan and wrong for the room. The best plan in the world adjusts on its feet. AI does not teach the lesson. You do. The plan is the preparation. The teaching is the work.

Final Verdict

Using AI to create lesson plans step by step works when the process is right. The tool is not the thing. The workflow is the thing. A strong prompt built from your own professional knowledge of your students, reviewed against a specific checklist, and personalised with what only you know, produces lesson plans that are faster than scratch built and often better than tired scratch built.

The Sunday afternoon where you sit in front of a blank document for twenty minutes does not have to be that way anymore. The blank document is the problem. The workflow in this guide is the solution to it.

Open the notebook first. Answer the four questions. Then open the tool. In that order. Every time.

The planning time you save goes back to your students in the way it always should have, as energy, presence, and the particular attention that only a teacher who is not exhausted can give.

That is the whole point of this workflow. Not efficiency for its own sake. Efficiency in service of the thing that actually matters.