The Feynman Prompt: How I Use ChatGPT to Learn Twice as Fast (5 Templates Inside)

Here is the problem with using ChatGPT as an answer machine: answers do not teach you anything. Understanding arrives when you try to explain something, fail, and repair the failure. The physicists call this the Feynman technique — explain the concept in plain language as if teaching a child, and the gaps in your explanation mark exactly where your understanding ends.

A chatbot cannot explain things for you, but it can be the tireless student who keeps asking you to try again. After a year of experimenting, I have settled on five prompts that turn ChatGPT or Claude into that student. Copy them, replace the bracketed parts, and keep them somewhere handy.

Working through a concept before checking the explanation

1. The Feynman Prompt

I am learning about [TOPIC]. I will explain it to you in my own words. Do not teach or correct me yet. After my explanation, ask me three pointed questions that would expose any gaps or misconceptions in what I said. Wait for my answers before moving on.

This is the core loop. The magic is in the instruction not to teach yet — you want the model to diagnose, not to rescue. My philosophy reading finally stuck when I had to explain Kant’s categorical imperative to a bot that kept asking, “But what if everyone lied?”

2. The Socratic Tutor

Act as a Socratic tutor for [TOPIC]. Guide me to understanding entirely through questions, one at a time, starting from what a beginner would know. Never give me the answer directly; if I am stuck twice on the same point, give the smallest possible hint.

Slower than reading an explanation, and that is the point. Retrieval effort is what builds memory. I use this one for concepts I will be examined on, and plain reading for concepts I just need to recognize.

3. The Misconception Hunter

What are the three most common misconceptions people have about [TOPIC]? For each one, explain why it feels right but is wrong, and give me a quick test question that would reveal if I hold it. Wait for my answers, then tell me which misconceptions I actually have.

Every field has its classic wrong beliefs, and knowing them in advance is like getting a map of the minefield. Claude in particular is good at the “why it feels right” part, which is where the real learning happens.

4. The Practice Exam Generator

Create a 10-question exam on [TOPIC] at [LEVEL, e.g. first-year university] difficulty: 4 recall questions, 4 application questions, 2 that connect the topic to something unexpected. Do not show answers until I have answered all 10. Grade me harshly and explain each mistake.

Testing yourself is the single best-studied learning technique in the literature, and this prompt industrializes it. The “connect to something unexpected” questions are worth keeping even if you skip the rest — transfer is what separates understanding from memorization.

5. The Gap Analyzer

Here are my notes on [TOPIC]: [PASTE NOTES]. Based only on what I wrote, list: (a) concepts I explained well, (b) concepts I mentioned but did not explain, and (c) important subtopics completely missing from my notes. Be specific and quote my own words back to me.

This one works beautifully inside NotebookLM or Claude Projects, where your whole note collection is in context. It turns “reviewing” from re-reading into an audit.

Notes and self-testing remain the backbone of the method

Where to start

Do not import all five at once. Take whatever you are learning this week, run it through the Feynman Prompt once, and notice how different it feels from asking for an explanation. If you like the result, add the Practice Exam Generator before your next test. The other three will earn their place as you need them. And if you want to automate the boring parts of this loop, my next post walks through the exact workflow I use.