Study prompt

Learn Anything Prompt — The Feynman Technique, Automated

The Feynman technique is the gold standard for learning: explain the concept simply, find the gaps, fill them, repeat. It’s effective and slow. This prompt automates the slow parts — generating the simple explanation, the harder example, the misconceptions, and the quiz — so you can run the loop in 10 minutes instead of an hour.

✓ Works in ChatGPT, Claude, Gemini ✓ Tested template ✓ Copy + paste ready

The template

Replace the [bracketed] fields with your specifics. Send to any AI chat.

I want to understand [concept] well enough to use it / talk about it intelligently in 10 minutes.

I already know: [what you know]
I don't know: [where you're starting]

Teach me with:
1. The core idea in 1 sentence (no jargon)
2. The simplest possible example
3. One slightly harder example showing where it gets interesting
4. The 3 most common misconceptions
5. The next thing I should learn after this

End by quizzing me with 3 questions, easiest first.

How to use it

Be specific about the concept. ‘Teach me machine learning’ is too broad; ‘Teach me how backpropagation works in a 2-layer neural network’ gets a useful answer.
Being honest about what you already know prevents the model from over- or under-explaining. ‘I know basic calculus and Python’ calibrates the response.
Actually answer the quiz questions, then ask ‘Did I get them right? Where am I confused?’ Repeat until the model says you’ve got it.

Example output

Here’s an example of what a quality AI returns when you run this template:

**Core idea:** Bayesian probability is updating your belief based on new evidence, in a math-formalized way. **Simplest example:** You think a coin is fair (50/50 heads). You flip it 5 times and get all heads. Your belief should now be that the coin is probably biased toward heads. Bayes’ theorem tells you exactly how much to update. **Slightly harder:** A test for a rare disease has 99% accuracy. You test positive. What’s the probability you have the disease? Intuition says ~99%. Bayes’ theorem says it could be as low as 9% if the disease is rare enough — because false positives from healthy people outnumber true positives from sick people. **3 common misconceptions:** 1. “If the test is 99% accurate, a positive result means I’m 99% likely sick.” Wrong — base rates matter enormously. 2. “Bayes is just a formula.” No — it’s a way of thinking about belief updates that survives even when you can’t compute the formula exactly. 3. “Frequentist vs Bayesian is a religious war.” Mostly a tooling preference at this point; modern stats practice uses both. **Next thing to learn:** Conjugate priors (a math trick that makes Bayesian updates computationally clean). — **Quiz:** 1. Easy: If your prior belief that an event will happen is 50%, and new evidence is 2x more likely under that hypothesis, what’s your posterior belief? 2. Medium: Why does a rare disease test with 99% accuracy still give you only ~9% probability of being sick after a positive test? 3. Hard: When would Bayesian inference give a different answer than maximum likelihood estimation, and why?

Why this prompt works

’10 minutes’ constrains the depth.

Without a time anchor, models default to either too-shallow (‘here are 5 facts about X’) or too-deep (‘let me start with the 1763 origin of the concept’). 10 minutes calibrates to ‘enough to talk about it intelligently’.

Two examples (simple + harder) is the secret to retention.

One example creates a false sense of understanding. Two examples — where the second introduces complexity — exposes whether you actually got the principle.

Misconceptions are the highest-information section.

What you DON’T know is harder to articulate than what you do. Asking for the 3 most common misconceptions is the cheapest way to find your own.

Quiz at the end converts passive reading into active recall.

Reading alone gives you a sense of understanding without testing it. Three questions at the end converts the session from ‘consumed content’ to ‘verified comprehension’.

Which AI to use

**Claude** for clear, pedagogical writing. **ChatGPT** for technical and STEM concepts where math reasoning helps. **Gemini** for concepts where its Workspace integration lets you connect learning to docs/notes you already have.

Read the full comparison in ChatGPT vs Claude vs Gemini in 2026 →

Frequently asked questions

How is this different from just asking ChatGPT ‘explain X’?

The Feynman structure forces a specific pedagogical shape: simple → example → harder example → misconceptions → quiz. Without that structure, you get a Wikipedia-style information dump that’s hard to retain.

Can I use this for technical topics like programming languages or math?

Yes — it works best on topics that have a ‘core idea’ you can articulate. For broad domains (‘teach me JavaScript’), narrow it down to one concept at a time (‘teach me how JavaScript’s event loop works’).

What if the AI’s explanation is wrong?

It happens, especially for niche or recent topics. Cross-check the core claim against a primary source (textbook, paper, vendor docs) before trusting it. The pedagogical structure is correct; the specific facts need verification.

Should I use this with the AI’s voice mode?

Yes — voice mode makes the ‘quiz me’ step feel more like real teaching. ChatGPT’s voice mode is the best for this; Claude’s is functional; Gemini’s has improved.

How do I know when I actually understand the concept?

When you can answer the AI’s hardest quiz question without re-reading, AND when you can give a simpler explanation than the AI gave you. The simpler your explanation, the deeper your understanding (this is Feynman’s actual insight).

Build any prompt in 30 seconds

Free tool, no signup, runs in your browser. Pick a category, fill in the blanks, copy your prompt.

Try the prompt generator → Read the full guide