Yuval Avidani
Author
You know this evening: it's ten PM, tomorrow there's a lesson on fractions, and your class has three levels: two kids already bored, four still struggling with the material, and everyone else somewhere in between. Building one lesson plan that fits all of them takes two hours you don't have. This is exactly where AI turns two hours into ten minutes, not to replace you, but to give you back the time to teach.
Wait, what AI am I even talking about here? When I say "AI" in this piece I mean a specific kind: a chatbot built on a language model, software you write a sentence to and it writes you back an answer, like a WhatsApp chat with an assistant. The familiar free examples: ChatGPT, Claude, Gemini. All you need to start right now is a browser and an account. Nothing to install, nothing to pay for, no prior technical knowledge required.
And why does this even work? Because a language model is trained on a massive amount of text from the internet, and what it knows how to do is guess the logical continuation of whatever you wrote. You type "build a lesson plan about...", and it continues the text the way a lesson plan would sound. This is the most important intuition here: the model doesn't "understand" the way you do, it completes text in a way that sounds right. This same principle is both its power, it can handle any topic in any format, and its danger: if something "sounds" right, it'll write it even when it's wrong.
The one thing separating a bad result from an excellent one: the context you give it
Before we get to any prompt (prompt = the instruction you write to the chatbot), let's understand why most people get mediocre results. They write a general request like "make me a lesson on the War of Independence." The model doesn't know what grade you teach, how much time you have, or who your students are, so it guesses at a gray average. The more detail you give, the less the model has to guess and the more precise it gets. That's the one rule to remember.
Step 1: A differentiated lesson plan in one prompt
"Differentiated" means: same topic, but adapted to three different student levels in the same class, so every kid gets a challenge sized just for them. Why does this matter? Because a struggling student who gets a question too hard gives up, and a top student who gets a question too easy gets bored, and either way you've lost them. Try this prompt (swap in your own details where the brackets are):
"You are an experienced history teacher for 7th grade. Build a 45-minute lesson plan on the establishment of the state. Include: an engaging 5-minute opener, a lesson body with 3 activities, and a summary. Add differentiated adaptation at 3 levels: for struggling students, grade-level, and for advanced students. Note how long each stage takes and what materials are needed."
Notice what we did here: we gave it a role ("experienced teacher"), an audience ("7th grade"), a length ("45 minutes"), a structure ("opener, body, summary") and a special requirement ("differentiated"). Every one of these details is another anchor point that narrows down the model's guessing.
Got a lesson plan? Now here's the most important part: this is a dialogue, not a one-time order. Keep talking to it in the same conversation: "shorten the opener," "add a pair activity," "my students love soccer, give an example from there." The model remembers everything said in the current conversation and adjusts accordingly.
Step 2: Worksheets and questions at three difficulty levels
This is where AI saves you the most time. One request generates an entire question bank:
"Create 12 questions on photosynthesis for 6th grade. Split them into 3 levels: 4 basic recall and knowledge questions, 4 comprehension and application questions, and 4 higher-order thinking and open-ended questions. Add a separate answer key at the end."
This structure, basics, application, higher-order thinking, isn't arbitrary. It's built on Bloom's Taxonomy, which is simply an accepted educational ladder that orders thinking from light to heavy: first remembering a fact, then understanding and using it, and finally analyzing and creating. Why build a worksheet along this ladder? Because a struggling student starts at the first level and feels success, and a top student jumps to the open-ended questions and doesn't get bored, all from the same page, the same lesson.
Practical formatting tip: ask the AI to return the content as a table or in Markdown. "Markdown" is a simple text format with headers and lists that pastes nicely into Google Docs or Word, a page ready to print. Want an elementary school version? Add "in simple language, short sentences, and with an emoji for each question."
Step 3: Personal feedback for every student, without sitting for hours
Writing thoughtful feedback for 30 tests is grunt work. Notice the distinction here: the AI doesn't grade for you, it phrases. You checked and decided what's right and what's wrong; it just helps dress that up in encouraging words. After you've graded, write:
"Write encouraging, constructive feedback for a student who made an error calculating the area of a rectangle and also forgot units of measurement. Use positive language, with a short explanation of what to fix, in a tone suitable for 5th grade."
Always review and edit the feedback before it goes out to the student. You know the kid, the AI doesn't. It gives you a draft; the pedagogical judgment always stays yours.
Ready-to-copy prompts
- High school, English: "Create a reading comprehension exercise for 10th grade at B1 level, a short text about artificial intelligence, 5 multiple-choice questions and 2 open-ended questions, with a glossary." (B1 = intermediate level per the European language framework, CEFR.)
- Middle school, Bible studies: "3 open discussion questions about the story of Joseph and his brothers that spark thinking about jealousy and forgiveness, suitable for a class discussion in 8th grade."
The honest conversation: wrong facts and "detecting" AI work
We have to talk about this without sugarcoating it. We already mentioned that a language model only completes text that sounds right, it doesn't check whether what it writes is true. This is where "hallucinations" come from: the model invents a fact, a date, or a quote with total confidence, simply because that's what "sounds" logical as the sentence's continuation. So always verify factual content (dates, quotes, numbers) before it reaches the classroom.
And regarding students, here's the point nobody says out loud: don't chase after "gotcha" moments, build assignments that are AI-resistant instead. Tools that claim to "detect" whether text was written by AI aren't reliable. Studies have found high error rates with them, and they mistakenly flag legitimate human writing too, especially from students whose first language isn't Hebrew or English. Even the makers of these tools ask that disciplinary decisions not be based on their score. A real kid who wrote it themselves could get punished because of an algorithm that's wrong.
So what does work? Assignments that are hard to fake: ask for intermediate drafts, an essay about a personal experience, a reflection (= a personal explanation of the learning process you went through), or an oral presentation of what was submitted. Warning signs for text that might be AI-written: overly polished, generic language that doesn't sound like the student, and a lack of personal examples. But these are cues to open a conversation, not proof.
Most important: talk openly with your students about when AI use is allowed and when it isn't, and teach them to use it as a thinking tool, not a shortcut. This is a life skill, no less important than the material itself. Take one prompt from this article, open a chatbot, and plan tomorrow's lesson right now. Let's fly high.