ChatGPT Lesson Plans: A Fast Plan for the Wrong Class

A ChatGPT lesson plan takes about forty seconds. Type "lesson plan, Grade 9, photosynthesis, 45 minutes", and a tidy plan appears: objectives, starter, activities, plenary. It looks like a plan. It reads like a plan. And in most cases it is a plan for a class that does not exist.

I build AI systems for a living, I used to teach, and I have children in school. So I want to explain, without hype in either direction, what is actually happening when ChatGPT writes a lesson plan, why the result is generic, and what to do about it.

What ChatGPT is doing when it writes a lesson plan

ChatGPT is a large language model (a program trained on a very large amount of text to predict the next word). When you ask for a lesson plan, it produces the most likely lesson plan given your words. "Most likely" is the key phrase. The output is a blend of the thousands of photosynthesis plans it has seen.

That is why the plan is good in a general way and wrong in every specific way. It does not know:

• which textbook your school uses, or what page 86 says;
• that eight of your students cannot yet balance a simple equation;
• that your lab is booked on Tuesdays, so the practical is impossible;
• that your class works in English with Arabic or Urdu as the stronger language;
• what happened in yesterday's lesson.

None of this is a flaw in the model. It is the absence of context (the information about your situation that the model has no way to see). A plan without context is a plan for the average class in the training data. Your class is not that class.

A fast plan for the wrong class is still the wrong plan

Here is the trap. Because the plan arrives fast and looks finished, the temptation is to teach it with light edits. The edits are where the time goes. A Grade 9 science teacher I know timed it: forty seconds to generate, twenty-five minutes to rewrite the activities for her actual room, and she still missed that the plan assumed a practical she could not run.

Writing from scratch would have taken her about the same twenty-five minutes, and the plan would have started from her class instead of being bent towards it. Speed only counts if the output is close to usable. For most ChatGPT for teachers use, the starting point is far away.

The parent's view

As a parent, I do not mind if my child's teacher uses AI to plan. I mind if the plan was made for a child who is not mine. The whole value of a teacher is that they teach the class they actually have. Any tool that pushes a teacher towards the average class is working against that, however polished the output.

A better ChatGPT lesson plan prompt

If you are going to use ChatGPT, give it the context in the prompt. Here is a template that produces something closer to usable:

You are planning a 45-minute Grade 9 biology lesson. Class: 30 students, bilingual (English and Arabic), eight students weak on the word equation from last lesson, two students who finish early. Topic: how light intensity affects the rate of photosynthesis. Source: the text I paste below from our textbook, pages 90 to 92. Objective: students can describe the pondweed experiment and explain the result. Format: starter linked to last lesson, one teacher demo, one paired task, exit ticket with three questions. Key terms in English and Arabic. Do not add a practical; the lab is unavailable. Plan only from the pasted text.

Then paste the pages. The plan will be better, because the model now has the context. Notice what you had to do: type the class description, paste the textbook, state the constraints, and remember all of it again next lesson. The chat window keeps none of it. Tomorrow you start from a blank box.

Where the chat window runs out

A chat is a good place to think and a poor place to keep things. Three problems appear by the second week.

Context does not persist. The model has a memory of the current conversation only. Your class description, your pages, your constraints must be supplied each time, or the plan drifts back to the average.

There is no record. When a head of department asks what a plan was built from, the honest answer is "a prompt I no longer have". The plan floats free of any source.

Nothing learns from the lesson. The plan does not know it was taught, let alone what went wrong. The next plan starts cold. A planner that asks "what landed?" after the lesson and carries the answer forward is a different kind of tool.

For a list of prompts that work in any tool, with the context each one needs, see AI for lesson planning: 12 prompts that work and the context they need. For the five-minute check to run on any generated plan, read AI lesson plan generator: how to get a plan your class can use.

How TeachBuddy AI helps

We built TeachBuddy AI to fix the context problem rather than the speed problem. The speed was never the issue. In the lesson plan generator you pick a classroom profile you described once, type the topic, and add sources by page range from your Resource Library, including scanned textbooks. The model sees your class and your pages, every time, without you retyping either. The plan records what it was built from.

The AI drafts; the teacher decides. Objectives are suggested and editable. After the lesson, a two-minute reflection proposes updates to the class profile and notes for the next lesson, and nothing is written until you approve it. The next plan starts warm, from the class as it is this week. Free for teachers.

Frequently asked questions

Can I use ChatGPT for lesson planning?

Yes, and many teachers do. Give it the class description, the source text and the constraints in the prompt, and check the output against your room. Expect to retype that context every time.

What is a good ChatGPT lesson plan prompt?

One that names the class, the source pages, the objective, the format, the language and what to leave out, like the template above. A prompt that only names the topic and grade produces a plan for an average class.

Is a ChatGPT lesson plan good enough to teach from?

Sometimes, after editing. The risk is not that it is bad. The risk is that it looks finished while being built for students you do not teach. Always check the activities against your room, your equipment and your students' prior knowledge.

If you would rather give the context once and plan from it every week, create a free TeachBuddy AI account and describe your first class.