Most teachers do not need more explanations. They need to know which parts of a topic a student cannot yet put into their own words. A feynman technique chatgpt prompt answers that with a fixed loop: plain language first, then questions about the explanation itself until the gaps appear.
The four steps below are that loop, the copy-ready prompt that runs it, and the signs that an explanation is still too easy to be useful. The same prompt works on a whiteboard, on a revision sheet, or as a self-check a student runs at home.
What the Feynman loop asks the AI to do
The classroom version of the technique is four steps. Pick one concept. Write the explanation as if to someone with no background. Mark every word or assumption you leaned on without defining. Rebuild the explanation without those crutches.
The last two steps are where a model earns its place. A student will not interrogate their own explanation honestly, because admitting a gap costs something. A model asks the diagnostic question without impatience, and keeps asking until the answer stops being a recitation.
That is the whole value here. It is not a better explanation from the model. It is a sharper audit of yours.
The copy-ready Feynman prompt
Act as a patient tutor. I am going to explain [concept] in plain language, as if to someone who has never studied [subject]. Do three things in order. First, read my explanation and list every word, phrase, or assumption that a beginner would not know, or that I used without defining it. Second, for each one, ask me a single question that would show whether I understand it or am only reciting it. Third, wait for my answers before you correct anything. After I answer: say which answers were accurate, which were vague, and which revealed a gap. For each gap, name the specific idea I skipped. Do not rewrite my explanation unless I ask. Do not praise it. Do not invent facts, dates, or technical terms. If I use a term wrongly, quote my sentence and ask me to restate it. If the explanation is genuinely clear to a beginner, say so plainly instead of inventing difficulty.
Fill in these details
- Concept: one idea, not a unit or a chapter.
- Subject: the context that decides what background you can assume.
- Level: who the explanation is aimed at, in one phrase.
- Rounds: how many question rounds you realistically have.
Delete the brackets you do not need, but keep the two refusal rules. They are what stop the model from rescuing a weak explanation with its own.
Push the explanation until the gaps show
1. Make the first pass shorter than feels respectful
Ask for a hundred words. Unconstrained, most people write three hundred, and padding hides the gap. A short answer forces one idea per sentence, which is exactly where the crutches appear.
2. Insist on examples instead of adjectives
When a student writes that plants make energy using sunlight, ask what happens and what would stop if it stopped. Vague verbs mark the thinnest part of an understanding. Ask the model to flag every verb the reader cannot picture.
3. Ask for the counter-case
Have the model ask what would have to be true for your explanation to be wrong. A student who can name the condition their explanation depends on understands the boundary, not just the sentence.
Worked example: photosynthesis for Grade 8
A student writes: plants make food using sunlight, carbon dioxide and water, and chlorophyll catches the light.
The prompt returns four flags. Chlorophyll is undefined. Make food hides what is actually produced. Nothing says where the water enters. No mechanism is named between the light and the sugar.
Each flag becomes one question, and the student revises. The second explanation names the stages and says where the water comes in. That revision, not the first answer, is the part worth keeping.
Turn one explanation into an ai learning prompt sequence
Once the gaps are listed, feed the list into the next round. An explain like i am five prompt is useful here, but only as a second pass, after a real explanation already exists. Its job is to test whether the plain version survives the simplification.
- Round one: the tutor prompt above, run on the student's own words.
- Round two: the same prompt, run on the corrected explanation.
- Round three: ask for the three questions you would ask a student who still gets it wrong.
The third round is the one most people skip, and it is the one that transfers. Asking for the questions forces you to turn your own understanding into an assessment, which is a different skill from understanding it.
I have revised my explanation of [concept]. Now act as the teacher running this lesson. Write the three questions you would ask to find out whether a student really has it, and for each question write what a wrong answer would sound like. Ask the questions one at a time and wait for my reply. Do not tell me whether I am right until I have finished all three. Do not add any content beyond [the specific scope of the lesson].
Troubleshooting an explanation that stays too easy
- It praises you first. Cut the instruction to correct, and add: do not comment on the quality of the writing, only on the understanding.
- It explains anyway. Add wait for my answer before continuing, and refuse an explanation offered in place of an answer.
- It flags style, not concept. Say ignore grammar, spelling, and word choice.
- It invents difficulty. Add if the explanation is clear to a beginner, say so and stop.
Common mistakes with Feynman prompts
- Running it on a whole chapter. One concept per loop, or the flags become a list nobody reads.
- Letting the model rewrite first. If it rewrites before you answer, you are reading its explanation, not finding your own gap.
- Skipping the second pass. The revision is the teaching moment, not the diagnosis.
- Using it as an assessment. It belongs in revision and self-check, not in graded work you did not witness.
Frequently asked questions
Does the AI judge whether a student really understands?
No. It compares the explanation against the concept you named and flags what is missing or undefined. That is a useful check, not a measure of understanding, and it should never carry a grade.
Is explain like I am five the same as the Feynman technique?
No. The first asks for a simpler register, which can produce a vague answer that merely sounds simple. The second asks for a simpler register plus an audit, which produces questions and a revision. Use the second when you want learning rather than a quick sanity check.
Can students run these prompts on their own?
Yes, with a checkpoint. Ask them to send you their first explanation before the second round, so you can see which gaps they found themselves and which ones they still cannot name.
Run one concept end to end this week
Take the idea you are about to teach anyway, and run the loop on your own explanation before the lesson. The places where you stall are usually the places where students stall. For the underlying structure, start with the introduction to prompt engineering.