Use cases
How to use an AI lesson plan generator for a real class
Any model will produce a lesson plan. The hard part is producing one that fits forty-three real minutes, the class that struggled with equivalent fractions on Tuesday, and the two packs of fraction tiles you actually own.
Try this exact task
“Plan tomorrow's 45-minute grade 5 math lesson on adding unlike fractions.”
The staged prompt makes OpenCraft collect the class context before it writes anything, and budget the minutes out loud. You read the whole prompt before you send it.
- Class profile and period length
- The standard text and last lesson's result
- Materials actually on hand
- Timed plan as an editable DOCX
- Check-for-understanding and exit ticket
- One scaffold, one extension, one short version
Use the complete prompt
Help me plan one lesson. Before writing anything, ask me for: grade and subject, class size, exact minutes in the period, where this sits in the unit, what we did last lesson and how it went, prior knowledge I can and cannot assume, groups needing a different entry point described by need and never by name, the standard text I will paste verbatim, the specific misconception I am targeting, and the materials I actually have. Then write a timed plan whose segment minutes add up to the real period length and show the running total. Include a student-facing objective, success criteria, teacher and student moves per segment, one mid-lesson check for understanding, one scaffold and one extension tied to the objective, an exit ticket with what each answer would tell me, a materials and prep list, and a shortened version for a cut period. Do not invent or paraphrase a standard code, and do not cite studies I have not given you. Flag any segment that assumes a material I did not list. Deliver it as an editable document.
The generic lesson plan is the failure mode
A plan with no class in it fits no class
Ask any model for “a lesson plan on photosynthesis” and you will get one. It will have objectives, a hook, guided practice, and a closing. It will also assume a period length nobody has, prior knowledge your students may not hold, a projector you may not own, and a pace that runs twenty minutes long. Nothing in it is wrong, exactly. It just was not written for anyone.
The U.S. Department of Education's 2023 report on AI in education describes this directly when it maps the “long tail of learner variability.” Standardized materials serve the middle of the distribution; most of the students who need a plan to be different are outside it. A lesson plan generator that knows nothing about your room is, by construction, teaching to the middle.
“AI can bring opportunities to address a wider spectrum of strengths and needs”
U.S. Department of Education, Office of Educational Technology
Four ways to spot a plan you cannot teach
Before you edit a draft, look for the tells. The timing does not add up to your period, or has no minutes at all. The materials list includes things you would have to buy or print with no printer. The differentiation is a sentence like “provide additional support to struggling learners,” which names no scaffold. And the check for understanding is a show of hands rather than a question whose wrong answers tell you something specific. Any one of those means the model was writing a template, not a plan.
The context that makes a plan usable
Five inputs do most of the work
You do not need a long brief. You need five things the model cannot guess, and they are the same five every time:
- The class profile. Grade, subject, size, and the groups that need a different entry point — described by need, never by name.
- The prior lesson and its result. Not “we covered equivalent fractions,” but “we covered equivalent fractions and about a third of them still added denominators.”
- The real minutes. Forty-three, not “a class period.” Say what the passing time, attendance, and packing up actually cost you.
- The materials on hand. Two packs of fraction tiles, a whiteboard, no 1:1 devices, printing available the day before.
- The specific misconception you are targeting. This is the single input that changes a plan the most, and the one people leave out.
Paste the standard; do not ask for it
Standards alignment is where these tools quietly fail. A model will produce a plausible-looking code and plausible-sounding wording, and the two may not match. Open your state or national standards document, copy the exact code and text, and paste it into the prompt as the target. For grade 5 fractions in the Common Core, that is 5.NF.A.1 — add and subtract fractions with unlike denominators by replacing them with equivalent fractions. Ten seconds of copying removes the entire class of error where a plan cites a standard it does not teach.
Ask for differentiation with a shape
“Differentiate this” produces filler. Name the shape you want instead: one scaffold and one extension, both tied to the same objective, both doable inside the same minutes. CAST's UDL Guidelines are a useful vocabulary here — multiple means of engagement, representation, and action and expression — because they push you to ask for a second route to the same target rather than an easier target.
What a usable plan must contain
A plan is a document you teach from, under time pressure, while twenty-eight people watch. Judge a generated draft against this list before you judge its prose.
| Component | What good looks like | The failure it prevents |
|---|---|---|
| Objective | One student-facing “I can” statement with success criteria | A lesson that covers a topic without teaching a skill |
| Standard | Code and wording copied from the official document | A fabricated or mismatched alignment |
| Timed sequence | Minutes per segment, summing to the real period | Running twenty minutes long and dropping the close |
| Check for understanding | A specific mid-lesson question with predicted wrong answers | Discovering the misconception at the exit ticket |
| Differentiation | One named scaffold and one named extension | “Support struggling learners” as a plan |
| Materials and prep | Only what you own, with what to do the night before | Arriving without the manipulatives the plan assumes |
| Assessment | Exit ticket plus what each answer would tell you | Collecting evidence you cannot act on tomorrow |
| Contingency | A shortened version for the day the period is cut | Improvising the assembly-day lesson at 8:55 |
Vague request versus specified request
The vague request
“Write a lesson plan for 5th grade math on adding fractions with unlike denominators.”
What comes back: a sixty-minute plan with a ten-minute warm-up, a fifteen-minute “I do,” group work at tables, a fraction-tile manipulative activity, an exit ticket, and a line about supporting struggling learners. Reasonable on paper. In the room it is fifteen minutes too long, the group work assumes pods rather than rows, and the tile activity assumes a set per pair when you own two packs.
The specified request
“Grade 5 math, 27 students, 43 usable minutes, desks in rows. Lesson 4 of 8. Last lesson was equivalent fractions on a number line; about a third still add denominators when the units differ. Target misconception: adding numerators and denominators straight across. Standard, verbatim: [pasted 5.NF.A.1 text]. Materials: whiteboard, two packs of fraction tiles, printing available yesterday only, no devices. Six students read well below grade level; four are multilingual learners at an emerging level. Give me a timed sequence that sums to 43, one scaffold, one extension, a mid-lesson check, an exit ticket with what each answer tells me, and a 28-minute version for assembly day.”
What changes: the tile activity becomes a single teacher-led demonstration at the board because two packs will not cover thirteen pairs. The warm-up becomes three equivalent-fraction items aimed at the third of the class that wobbled, so you find out in minute four rather than minute forty. The check for understanding becomes one deliberately chosen item where the wrong answer is the straight-across sum, which makes the misconception visible instead of inferred. The extension is a same-objective problem with three addends rather than a different topic. And the close is real, because eight minutes were reserved for it before the middle was written.
The second prompt is not more clever. It is the same request with the five inputs attached. That is the whole technique.
Keep the plan and its materials in one thread
What OpenCraft actually produces
OpenCraft writes the plan as a real document, not a wall of chat text. Its document tool takes Markdown and renders an editable .docx alongside a styled HTML preview, so you open the plan in Word or Google Docs and edit it like any other file. Because the conversation keeps a persistent workspace, the same thread can then build the collateral the plan calls for — slides for the board, or a handout — from the plan you just approved rather than from a fresh, slightly different idea of the lesson.
That consistency is the point. Slides are generated by running real code in a sandbox that produces a .pptx, renders it to PDF, and checks the result before anything is handed back, so a broken deck fails rather than arriving quietly broken. Files with the same name stack as versions, which means next year's revision of the plan is a new version of one document instead of the fourth file called fractions-lesson-final.
What it does not do
It does not know your students. It has no connection to your gradebook, your LMS, or your roster, and it should not have one for this task. It cannot verify that a standard code is current for your state — that is why you paste the text. It does not observe whether the pacing worked, so the second draft is only as good as what you tell it after teaching the first. You can upload a curriculum map or a textbook chapter as a PDF and it will read the text, and it can search the web for a source, but neither substitutes for you checking the standard against the official document.
Different models handle long unit documents and tight instructions differently; you can compare models before committing a whole unit to one. When you need the student-facing practice items and answer key that the plan points at, that is a separate job — see the AI worksheet generator workflow.
Judgement, school policy, and student privacy
Timing and differentiation are judgement calls
A generated plan can hold minutes and structure. It cannot know that your fourth period needs the warm-up shortened because they arrive from lunch, or that one student will disengage if called on cold, or that the class has done three group activities this week and needs something else. Treat the draft as a starting structure you overrule. The federal AI report is blunt about this: teachers need to be able to see automated recommendations, disagree with them, and override them without penalty. Your plan, your call.
Check your school's policy before you paste anything
Many districts maintain an approved-tools list, a data-privacy review process, and rules about AI-assisted materials. Some require a disclosure when generated content is used with students. Find out before, not after.
Keep identifiable student information out of the prompt entirely. In the United States, student education records are protected under FERPA, which governs personally identifiable information from those records; other jurisdictions have their own equivalents. Names, student numbers, IEP or 504 documents, assessment scores, and behaviour notes do not belong in a general-purpose model. Describe groups by need instead — “four multilingual learners at an emerging level,” not four names — which is also, conveniently, the description the plan actually needs.
For other working documents built the same way, browse the OpenCraft use-case library.
Sources
- Artificial Intelligence and the Future of Teaching and Learning: Insights and Recommendations U.S. Department of Education, Office of Educational Technology
- Universal Design for Learning Guidelines, version 3.0 CAST
- Grade 5, Number and Operations—Fractions, standard 5.NF.A.1 Common Core State Standards Initiative
- What is FERPA? U.S. Department of Education, Student Privacy Policy Office
Frequently asked questions
What is an AI lesson plan generator?
It is software that turns a description of a class, an objective, and a set of constraints into a structured teaching plan: objectives, a timed sequence, materials, checks for understanding, differentiation, and an assessment. It writes the teacher's planning document, not the students' practice materials.
Are AI-generated lesson plans any good?
They are as good as the context you supply. With a one-line request you get a plan written for an average class that does not exist, usually over-scoped for the period. With a class profile, the real period length, last lesson's result, and the materials you own, the draft is close enough to be worth editing.
Can an AI lesson plan generator align to standards?
It can format a plan against a standard, but it cannot be trusted to recall a standard code correctly. Paste the exact code and wording from your state or national standards document into the prompt, and check any code the model produces on its own against the official source before you put it in front of an administrator.
How long does it take to make a lesson plan with AI?
Writing the brief takes about ten minutes the first time and two or three minutes after that, because most of it is stable across a unit. Generating and revising the plan takes a few minutes more. The saving is in the structure, timing table, and materials list, not in the thinking about your students.
Can I edit an AI lesson plan in Word or Google Docs?
Yes, if the tool gives you a DOCX rather than a block of chat text. Ask for the editable document up front, revise it there, and keep it as the file you reopen next year, because the plan is a document you will amend rather than regenerate.
Is it safe to put student information into an AI tool?
Do not put names, student IDs, IEP documents, grades, or anything else that identifies a student into a general-purpose model. Describe groups by need instead. Many districts have an approved-tools list and a data-privacy review, and student records are protected under laws such as FERPA in the United States, so check your school's policy first.