What the Feynman technique is
The Feynman technique is a way to test whether you understand something by explaining it in plain words. It's named after the physicist Richard Feynman, who was known for explaining hard physics simply. He didn't write it down as a four-step method himself. Others attached his name to it later, based on how he described learning. The steps are still useful whatever their origin:
- Pick a concept and write its name at the top of a page.
- Explain it in simple words, as if you were teaching someone who's never heard of it.
- Find the places where your explanation got vague, skipped a step, or fell back on jargon. Go back to the source and fill those gaps.
- Simplify again, and add an example or an analogy.
The method works because explaining forces you to put ideas in order and connect them. You can't fake that with recognition. Plenty of things feel understood until you try to say them out loud.
Where it goes wrong when you do it alone
Step 3 is the hard one. When you explain something to yourself, you fill your own gaps without noticing. You know what you meant, so the jump from A to C feels fine. A real listener would stop you and ask about B.
That's the job ChatGPT can do. It reads exactly what you wrote and can ask about every step you skipped.
The big mistake: letting ChatGPT do step 2
The obvious way to "use ChatGPT for the Feynman technique" is to ask it to explain the topic simply. That's ELI5, and it's useful, but it's not this technique. If ChatGPT does the explaining, you've done some reading. You haven't tested anything.
In the Feynman technique, you explain and ChatGPT listens. Set that up clearly at the start.
How to run it with ChatGPT
Step 1: Set the role
I'm going to explain concept to you as if you're a smart 15-year-old who's never studied it. Don't explain it to me. After my explanation, ask me questions about every step that's unclear, any word I didn't define, and anything that sounds hand-wavy. Ask one question at a time.
The "smart 15-year-old" framing works better than "a child". You want a listener who'll follow a real argument but won't accept "it just does".
Step 2: Explain without notes
Type or dictate your explanation. Don't look anything up. Voice mode suits this step, because talking reveals the gaps faster than typing does. You can't quietly reword a sentence halfway through.
Keep it short at first. Three or four sentences is enough for a first pass. You'll add detail as the questions come.
Step 3: Answer the questions
This is the part that matters. Some questions you'll answer easily. Some will make you realize you've been repeating a phrase you never understood. Those are what you came for.
When you hit a gap, don't ask ChatGPT for the answer straight away. Go back to your notes or textbook first, find the missing piece, and then answer the question in your own words. What you find yourself usually sticks better than what you're told.
Step 4: Ask for a gap report
Summarize the gaps in my explanation: which steps I skipped, which terms I used without defining, and anything I said that was wrong.
Save that list. It's a far better revision guide than rereading the chapter.
Step 5: Explain it again, shorter
Here's my second attempt. Is it accurate? Is anything still vague?
The second explanation is usually shorter and clearer. That isn't you dumbing it down. You now know which parts carry the idea.
A short example
Say you're explaining compound interest. Your first attempt:
Compound interest is when you earn interest on your interest, so your money grows faster over time.
A good follow-up from ChatGPT in its listener role would be: "Faster than what? And when does the interest get added? Every day, or once a year?"
That's the gap. You know the phrase "interest on your interest", but you haven't said what it's compared with or why the timing matters. Your second attempt might be:
With simple interest, you only earn interest on the money you put in. With compound interest, the interest you've earned gets added to the balance, and the next round of interest is calculated on that bigger balance. So each period, you earn a bit more than the last. The more often it's added, the faster the balance grows.
Now try it with numbers. "Put 1,000 in at 5% a year, compounded yearly. What's the balance after two years?" If you can work out 1,102.50, you've got it.
Prompts to copy
- Listener mode. "I'll explain topic. You're a curious student who doesn't know it. Ask me about anything unclear, one question at a time. Don't explain it yourself."
- Tougher listener. "Now be a skeptical expert. Point out anything that's technically wrong or oversimplified."
- Jargon check. "List every technical term I used and ask me to define each one in plain words."
- Transfer test. "Give me a situation I haven't seen and ask me to use the concept to predict what happens."
- Gap report. "What were the weakest parts of my explanation?"
When you're starting from zero
The Feynman technique assumes you've met the idea at least once. If you haven't, you'll have nothing to explain, and the session turns into ChatGPT teaching you after all.
For a brand new topic, get a first explanation, then do the Feynman pass a day later. One way is to ask ChatGPT for a short narrated lesson with Lesson Video. It plays in the chat with captions and a question at the end. Once you've watched it, close it and explain the idea back to ChatGPT using the prompts above.
For a lighter version of this, the explain like I'm 5 guide covers getting good simple explanations. For drilling the details afterwards, see active recall with ChatGPT.
Why it's worth the discomfort
The first round of questions usually feels bad. You find out that a topic you thought you knew has holes in it. That's the point. The holes were there before. Better to find them now than the night before the exam.