How to improve AI prompt for lesson planning: a repair method for teachers

Improve AI prompt for lesson planning by stating the grade and class, the learning goal, the time available, and the limits you need, then asking for a specific output format and a list of points to verify before class. A vague request like “make a lesson on fractions” forces the model to guess all of these, so the result often looks plausible but doesn’t fit your students. Checking each element in order helps make a weak prompt more usable.

Start With the Prompt You Already Have

You do not need a new habit to start. You need a way to see what your current prompt leaves out. Begin with a prompt you have actually typed:

Make a lesson on fractions.

The output is usually a tidy outline: a warm-up, a short explanation, a worksheet, and an exit ticket. It reads well. The problem is what it does not say. It does not know your grade, whether students have seen equivalent fractions, how long the period is, or what you cannot do with the class.

Read the output for what it is missing

Go through the draft and ask five questions. Who is this for? What should students be able to do at the end? How much time does it assume? What did it assume about materials or reading level? What should I check before I teach it? Each question you cannot answer from the draft points to a missing element in your prompt.

Why a one-line request produces generic plans

When a model has no details, it fills the gaps with the most common version of a lesson. That version is written for an average class that does not exist. The plan may look complete while still pacing a 45-minute period as if it were 90 minutes, or using a grocery-store word problem your students have never encountered. The fix starts with noticing what the model had to invent.

Add the Four Things a Lesson Prompt Needs

A lesson prompt needs four elements: context, a learning goal, constraints, and a requested format. The goal is one sentence describing what students will do at the end, such as “Students explain why 1/2 and 2/4 name the same amount.” The format is a request for a table, numbered steps, or timings. Context and constraints need more detail, so each has its own checklist below.

Context: grade, class, and prior knowledge

• Grade level, for example “Grade 6”
• Class size and mix, for example “28 students, several need reading support”
• What students already know, in one or two sentences
• Reading level of any text the lesson includes

Compare “make a fractions lesson” with “Grade 6, 28 students, who can add fractions with like denominators but have not compared unlike ones.” The second version changes the output. The model drops the basic addition warm-up and starts with comparing two fraction strips.

Constraints: time, standards, and what to avoid

• Class length, including the exit ticket, for example “50 minutes”
• The topic or standard your school uses, named by you
• Materials you actually have, for example “whiteboard and paper strips, no devices”
• Things the lesson must not include, for example “no timed quizzes, no groups larger than four, nothing outside this unit”

The avoid list matters as much as the rest. Without it, the model adds activities you cannot run, and you spend time cutting them out.

Ask the Model to Show Its Checks

End the prompt by asking for verification material. Request three things: a list of assumptions the model made about your class, claims or facts you should confirm, and questions students commonly get wrong, with the reason. A line like this works well:

End with three sections: (1) assumptions you made about my class, (2) facts or examples I should confirm, (3) two questions students commonly get wrong and why.

These checks help your review. They do not replace it. AI output is not guaranteed to be correct, and a list of assumptions only shows you where to look.

A Before-and-After Example

Start with a weak prompt for a middle school math lesson: “Make a math lesson on ratios for middle school.” The output is a generic plan with a recipe word problem, a worksheet of ten items, and no timing. Now rebuild it, labeling each element:

[Context] Grade 7 math, 26 students, mixed reading levels. They can multiply and divide whole numbers and know what a fraction is. [Goal] Students decide which of two lemonade recipes tastes more lemony, using ratios. [Constraints] 50 minutes, whiteboard and printed cards only, no calculators, no percent work. [Format] A timed agenda table, then three discussion questions. [Checks] List your assumptions, facts to confirm, and two likely student errors.

Illustrative example only, not real output from Toula Studio or a tested prompt.

What changed	Weak prompt output	Repaired prompt output
Warm-up	Generic definition of a ratio	Quick question that uses fractions students already know
Timing	None	Agenda that adds up to 50 minutes
Errors	None listed	Confusing part-to-part with part-to-whole

The teacher still needs to check the new output. Confirm that the recipe numbers are correct, that the timing matches your routine, that the topic fits your unit, and that the listed errors match what your students actually do. The prompt narrowed the guesses; it did not remove your review. Toula Studio also lets you practice finding errors in a deliberately flawed AI-written lesson plan or grading rubric, then get written feedback on your checks.

Where a Repair Tool Helps and Where Your Judgment Still Matters

The checklist works without any tool. Use it on paper or in a note before you type. A prompt-scoring tool can speed up the first pass, though. Prompt Repair Studio in Toula Studio scores a prompt for clarity, context, constraints, and verification, then offers a rewritten version. Use the score to find the weakest element first, then compare it with your own reasoning.

The rewrite still needs teacher review against your class. A cleaner prompt can still produce content that is wrong or does not fit your students, and Toula Studio does not guarantee correct content.

Frequently Asked Questions

Does a longer prompt always produce a better lesson plan?

No. Length helps only when each added sentence fills a real gap, such as grade, goal, time, or limits. Padding with vague praise like “make it engaging” adds words without information. Check whether each sentence answers one of the four elements.

Should I ask the AI to check its own lesson plan?

Yes, but treat the result as a list of places to look, not as a verdict. Ask it to list its assumptions and the facts you should confirm. Then check those yourself against your curriculum and your students.

Can a tool rewrite my prompt for me?

Prompt Repair Studio can score your prompt and offer a rewritten version. The rewrite is a starting point. You still need to confirm that it matches your class, your time, and your standards before you use the lesson.

Try this with one of your own lesson-planning prompts. Paste it into Toula Studio and open Prompt Repair Studio to see the score for each element. Then compare the rewrite with your original. The checklist in this article works without the tool, so you can use it either way.