The Feynman Technique for the AI Era: Catch What You Don't Understand

The Feynman technique is a study method: explain a concept in plain words, find where your explanation breaks, relearn that piece from the source, and explain it again. Aim the explanation at someone who has never met the idea, so jargon can't cover the gaps. It works because producing an explanation exposes what rereading hides, and with AI you can add a listener that questions every vague step before it explains anything.

This page covers the four steps of the Feynman technique, a prompt that turns an AI chatbot into the skeptical listener, an example round on electromagnetic induction, a checklist for spotting gaps, and the cases where the method is the wrong tool.

A physics student leaves a lecture on electromagnetic induction sure she understands Faraday's law: she can restate it and nod through the worked examples. Two weeks later a problem set asks why a bar magnet falls more slowly through a copper tube than through an identical plastic one, and she freezes. She can recite the law. She can't use it on a case she hasn't seen.

The gap between "I understand this" and "I can produce an answer from this" has a name, and it is worse than plain ignorance, because it feels like knowledge. You find out only when a question makes you build the answer yourself, usually in the exam.

Why does explaining something expose what you don't know?

Psychologists call it the illusion of explanatory depth. In a series of studies, Leonid Rozenblit and Frank Keil (2002) asked people to rate how well they understood how everyday things work, such as a zipper or a flush toilet. Then they asked them to write out the explanation step by step, and to rate their understanding again. The ratings dropped. People had confused knowing what a thing does with knowing how it works, and only the attempt to explain it showed them the difference. The illusion was strongest for explanations, and much weaker for facts, procedures and stories.

You can run their test on yourself in a minute. How well do you understand how a zipper works? Now, before reading on, explain step by step how the slider locks two rows of teeth together. If you got as far as "it pushes them together" and ran out, that is the illusion, caught in the act.

The Feynman technique gets around a bootstrap problem. To notice that you are wrong about something, you usually need to understand it well enough to catch the error, which is exactly what you lack. You don't need to know where the gap is in order to find it. You only need to try to produce the explanation. The stumbles, the hand-waves and the "and then something happens" moments surface the gaps on their own, and you can hear them yourself, as long as you are producing rather than rereading.

The research on explaining to yourself points the same way. In an early study of students learning mechanics from worked examples, Michelene Chi and colleagues (1989) found that the students who went on to solve problems well had generated more explanations of the example steps as they studied. It was a small study of eight students, and it showed a link, not a cause. Later experiments prompted students to self-explain on purpose, and a meta-analysis by Kiran Bisra and colleagues (2018), pooling 64 research reports, found a reliable average benefit (g = 0.55) for learners who were prompted to explain over those who were not.

The specific mechanism

When you read a well-written explanation, your brain matches it to what you already know, registers "this makes sense," and moves on. Little new is built, because nothing had to be produced. When you explain the same concept from scratch, you have to rebuild the causal chain, and every missing link forces a stop. You can smooth over a gap you are reading. You cannot smooth over a gap you are producing. So the place where your explanation stalls is useful: it marks exactly what to study next.

That is why the "as if to a child" part matters. It strips out the technical vocabulary that experts and students alike use to cover gaps. If you can only explain Faraday's law with the phrase "flux through the surface," you have memorized a formula and its label. If you can say in plain words why a moving magnet makes a current flow, you understand it. The beginner audience forces the stripping.

Did Feynman invent the Feynman technique?

Not in this form. Richard Feynman, the Nobel Prize-winning physicist known for his lectures to undergraduates, never wrote down a four-step method. The name is a later label for habits his colleagues and biographers describe. The best-known one comes from James Gleick's biography (1992): while preparing for his oral qualifying exam in graduate school, Feynman opened a fresh notebook, titled it "Notebook Of Things I Don't Know About," and spent weeks taking apart each branch of physics he knew, looking for the raw edges and inconsistencies. The four steps below are the modern summary of that habit and of his insistence on plain explanation.

How do you use the Feynman technique with AI?

The four steps, with the prompt that makes an AI chatbot useful for them. One thing Feynman's notebook could not do was ask questions back. A chatbot can, and it never gets tired of playing the beginner.

Step 1: Pick one concept

Narrow enough that you could explain it in a single paragraph if you understood it. Not "electromagnetism." Instead: "Why does a changing magnetic field induce a current in a wire?" Or: "Why is the derivative of $x^2$ equal to $2x$?" Or: "Why is the money you didn't earn part of the cost of a choice?" You are done when the concept fits in one question that starts with Why or How.

Step 2: Explain it, with the source closed

Write it in a text file or say it into a voice note. The one rule: you cannot look at the source material. Use only what is already in your head. Take the explanation end to end, with no "I'll come back to this part." Fill every gap with something, even if the something is "I think this happens, but I'm not sure why." Those sentences are the ones you are looking for. You are done with this step when the explanation runs from start to finish, however rough.

Step 3: Have the AI play the curious skeptic

This is the step that used to need a patient friend. Paste your explanation with this prompt:

Below is my explanation of {CONCEPT}. Play the role of a sharp, curious student
who takes nothing for granted. For every sentence I write:

1. If something is vague or hand-waved, ask me to be specific.
2. If I use a technical term without defining it, ask me to define it.
3. If I make a claim that has exceptions, edge cases, or hidden assumptions, probe them.
4. If my explanation has a logical gap — a step that does not follow from the previous —
   point it out and ask me to fill it.

Do NOT give me the correct explanation yet. Your job is to find every weakness and ask me
to defend it. Go through the whole thing.

My explanation:

{YOUR EXPLANATION}

Expect a long list of questions. Some will feel pedantic, and that is fine. Each one points at a place where your explanation went soft and you smoothed over it. You are done when you have the full list, unanswered.

Step 4: Answer each question, then loop

Try to answer every question, still without the source. The ones you can answer are not gaps. The ones you can't are. For each real gap, go to the source, learn that specific piece, and fold it back into your explanation. Then rewrite the whole explanation and run Step 3 again.

Stop when a round turns up no question you can't answer, and when you could hand your explanation to a classmate who missed the lecture. A few days later, explain it again from scratch with the book closed; if it still holds, move on. By then the explanation is tight in a way rereading can't make it, because rereading never forced you to produce anything.

What does a Feynman round look like on a real concept?

Here is Step 3 on the induction example, written for this page. A student has explained electromagnetic induction in free text, and the AI plays the skeptic.

Student's explanation (abbreviated): "When a magnet moves near a wire, it creates an electric current in the wire. This is called induction. The strength of the current depends on how fast the magnet is moving and how many coils the wire has. The current flows in a direction that opposes the change in the magnetic field, which is Lenz's law."

AI: "Questions:

1. You said 'when a magnet moves near a wire it creates an electric current.' Does the magnet itself create the current, or is something else doing it? What exactly is moving inside the wire when current flows?
2. You used the word 'field' without defining it. What is a magnetic field?
3. You said the strength depends on 'how fast the magnet is moving.' Is it the speed that matters, or the rate of change of something else? If I moved the loop at constant speed through a region where the field was the same everywhere, would a current still flow?
4. You said the current 'opposes the change.' Opposes it how? What would go wrong if it helped the change instead?
5. Does the wire have to form a closed loop for a current to flow? Would a straight, open-ended wire do the same thing?"

Before you read on, try question 3 yourself.

It is the heart of the concept: induction is driven by the rate of change of magnetic flux through the loop, not by motion as such. In a uniform field the flux through the loop doesn't change, so no current flows, however fast the loop moves. The student had memorized "faster magnet, stronger current" as a pairing, not as a consequence of $\frac{d\Phi}{dt}$. Question 4 leads to the same place from the other side: if the induced current helped the change, it would feed itself energy from nowhere, which is why Lenz's law follows from energy conservation.

The student goes back to the source, learns the flux formulation, answers the questions and rewrites the explanation. Round two reads: "A current flows when the magnetic flux through a closed loop changes, and how fast it changes sets the strength. The current's own field pushes against that change, because if it helped, the loop would make energy from nothing." The skeptic's next pass finds far less to ask.

Now the copper tube is easy. The falling magnet changes the flux through each ring of copper below and above it, the currents induced in the copper oppose that change, and the opposition slows the magnet. Plastic carries no current, so nothing pushes back. The problem that froze the student at the start is answered by her own second-round explanation, not by a formula she memorized.

How can you tell your explanation has a gap?

Run each draft of your explanation against this checklist:

1. Did I use a technical term without defining it? If yes, I don't understand that term yet. Find the definition.
2. Did I make a claim without a mechanism? "X causes Y" without saying how. If yes, I don't understand the cause.
3. Did I use hedging words to cover uncertainty? "Somehow," "kind of," "in some way." If yes, the gap is right there.
4. Did my explanation lean on a concept I haven't explained? If yes, that concept is a prerequisite I may not understand either.
5. Could someone with no background solve a related problem after reading my explanation? If no, the explanation is incomplete.

Any "yes" to questions 1 to 4, or a "no" to question 5, is a gap worth fixing before you move on.

When is the Feynman technique the wrong tool?

Two cases where the loop doesn't fit:

1. Pure memorization. The symbol for potassium, the date of the Treaty of Westphalia, irregular Spanish verbs. There is no mechanism to explain: the fact is the fact. Use active recall and spaced repetition for these instead.
2. Skills, not concepts. Riding a bicycle, playing the violin, writing idiomatic Rust. These are procedures, not explanations. Describing how you do them changes little; deliberate practice on real problems is what moves you forward.

The Feynman technique is for understanding concepts: the why and the how. That covers much of what school tests, and most of what breaks down later when you try to use what you learned somewhere new. It is also where the illusion of understanding does the most damage.

How do you run the Feynman loop in Ritsu?

The prompt above works, but you are holding it together. A chatbot tends to drift back into explaining a few messages after you told it not to, it judges your explanation against whatever it knows rather than your textbook, and nothing brings the shaky concept back next week. Keeping the skeptic honest is on you, message after message.

Ritsu runs the same loop on your own file. It asks before it explains, grades what you wrote against the material, and tells you what you got right and what you missed. The video at the top of this page shows it on a psychology chapter about memory, a different subject from the induction example above, but the same loop.

The moment that matters is the verdict. In the video, a practice case about a commuter named Maya is answered with the two kinds of mental processing swapped. The feedback names both problems in one sentence: "You have reversed the definitions of automatic and effortful processing and missed the core concept of how skills transition between the two over time." The second half is the gap a self-check misses: the answer never mentioned that effortful work turns automatic with practice, so nothing in it could have flagged the omission. The worked solution stays behind Show Solution until you ask for it.

1. Pick the concept: drop in the chapter. Choose File, add the PDF and press Start Learning. The chapter comes back as an ordered path of small units, each traced to the page it came from, so the concept you pick is already narrow.

2. Explain it before you read on. In the video, the lesson opens with one short question to see where you stand. Practice cases ask you to explain why, in your own words, and the solution stays closed until you have written an attempt.

3. Let it play the skeptic. Your answer gets the verdict above: what was wrong and what was missing, before any model answer.

4. Loop: retry, then let it come back. After a miss, Try a similar exercise brings a new case to explain. Press Add to Review and the session returns after 1, 3, 7, 16 and 35 days, so you explain it again while it is still shaky.

For steps 2 and 3 on one concept at a time, Teach Me The Concepts is the Feynman technique as a tool. You explain a concept from your material in a couple of sentences, each answer is scored on accuracy, coverage, depth and clarity, each out of 5, and the next question goes after the part you skipped. The induction question from this page is a fair first test. Every question and explanation in Ritsu shows the page of your file it came from, so the source stays the referee when you and the skeptic disagree.

Try it on the concept you'd least like to be asked about this week: upload the chapter, answer the first question with the book closed, and read the verdict before the model answer. Start Learning Free

Free forever · No credit card · No time limit

FAQ

Q: Is the Feynman technique effective?

Its parts are well studied. Prompting learners to explain material to themselves improves learning on average (Bisra and colleagues, 2018), and trying to explain something deflates a false sense of understanding (Rozenblit and Keil, 2002). Studies of the full named routine are scarce, so treat it as a sound method rather than a tested program.

Q: How long should a Feynman session take?

Give one concept one sitting. The first explanation takes a few minutes; most of the time goes into the gaps it turns up and the rounds after. If a concept keeps sending you back to prerequisites, split it and run the loop on the prerequisite first. It is a tool for understanding deeply, not for a quick review.

Q: Can I do the Feynman technique without writing anything down?

You can talk it through out loud, which still forces you to produce. Writing is stricter: a written sentence can't trail off into "you know what I mean," and you can hand it to a questioner line by line. Rehearsing silently in your head is the weakest version, because it lets you skip the steps you can't say.

Q: What if the AI's questions are pedantic or wrong?

Some will be. The signal is the questions you can't answer. If you can answer one in a sentence, it isn't a gap. If the AI objects to something you got right, check the source and hold your ground. If you can't defend the claim, the gap is real even when the question was clumsy.

Q: Does the Feynman technique work in a study group?

Yes. The classic form is one person explaining while another asks the probing questions, then swapping. Classmates catch gaps an AI may miss, because they know what your course expects. An AI is the fallback when nobody shares your topic or your schedule, and it is there whenever you are.

Q: Can I use ChatGPT for the Feynman technique?

Yes, if you stop it from explaining first. Paste your explanation with a prompt that makes it play a skeptical student, question vague steps and undefined terms, and hold back the correct explanation. Check its objections against your textbook, since a chatbot can sound certain and still be wrong.

The takeaway

If you can't explain a concept without the jargon that paints over your uncertainty, you don't understand it yet, and the fastest way to find out is to try. For the question-first version of the same idea, read why answer-first AI tools hurt learning.

Pick the concept you'd least like to be asked about, explain it in plain words with the book closed, and let Ritsu find the gap in your own chapter. Start Learning Free

Sources

• Bisra, K., Liu, Q., Nesbit, J. C., Salimi, F., & Winne, P. H. (2018). Inducing self-explanation: A meta-analysis. Educational Psychology Review, 30(3), 703–725. https://doi.org/10.1007/s10648-018-9434-x
• Chi, M. T. H., Bassok, M., Lewis, M. W., Reimann, P., & Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. Cognitive Science, 13(2), 145–182. https://doi.org/10.1207/s15516709cog1302_1
• Gleick, J. (1992). Genius: The Life and Science of Richard Feynman. Pantheon Books. https://openlibrary.org/isbn/0679408363
• Rozenblit, L., & Keil, F. (2002). The misunderstood limits of folk science: An illusion of explanatory depth. Cognitive Science, 26(5), 521–562. https://doi.org/10.1207/s15516709cog2605_1

By Ritsu Team · Last updated October 19, 2026