Most “explain it simply” prompts produce explanations that are confidently wrong: the model lops off the inconvenient parts and you walk away with a clean misunderstanding. These 12 prompts force layered simplification — a true-to-the-theory short version, a tiered analogy, a worked example, and an explicit “what you should NOT take away from this” appendix — so the simplification never quietly turns into misinformation. Pair them with concept comparison prompts when you need to keep two close ideas straight.
TL;DR
- Each prompt below is copy-ready. Replace the bracketed
[theory],[problem], and[known theory]placeholders with your own, then paste. - The set is built so simplification and accuracy travel together: prompts 1, 3, and 6 simplify; prompts 5, 7, and 12 are the guardrails that catch the lies a simplification introduces.
- For the explanation work itself, Claude (Opus 4.7 or Sonnet 4.6) and ChatGPT (GPT-5.5, Thinking mode) are the strongest. See Which model to use below.
Best for
- Self-learning a hard subject (stats, ML, economics, physics, law)
- Tutoring and teaching prep
- Writing explainer articles or course material
- Onboarding a new teammate into an unfamiliar domain
- Exam revision where surface understanding has already failed once
Which model to use
As of June 2026, three families do this well, with real differences worth knowing:
| Model | Strength for explanations | Note |
|---|---|---|
| Claude Opus 4.7 / Sonnet 4.6 | Best at showing reasoning step by step and flagging its own uncertainty | 1M-token context; ideal for prompts 5, 7, and 12 (the accuracy guards) |
| ChatGPT GPT-5.5 (Thinking) | Strong structure and worked examples; picker has Instant/Thinking/Pro | Use Thinking mode for the multi-step prompts; Free tier limits are tight |
| Gemini 3.1 Pro | Live search, so it pulls current real-world facts into examples | Best when the theory’s “consequences” need up-to-date data (prompt 9) |
For the misinformation-guard prompts especially, Claude’s habit of saying “I’m not certain about X” is an asset — that honesty is the whole point of prompt 12. Pricing as of June 2026: Claude Pro $20/mo, ChatGPT Plus $20/mo, Google AI Pro $19.99/mo all unlock the stronger models. The free tiers run the explainers but throttle long sessions.
How to use these
- Pick the prompt that matches your failure mode (no intuition? use #3; can’t apply it? use #4; over-applying it? use #5).
- Swap the
[bracketed]placeholders for your real theory and problem. - Run prompt #12 last, feeding back the explanation you got, to scrub the simplification of confident errors.
1. Feynman-style 5-step explainer
Explain [theory] using Feynman's method. Output 5 steps: (1) name the theory, (2) explain to a 12-year-old, (3) point out where the simple version is wrong, (4) introduce 1 added nuance, (5) the real definition.2. First-principles breakdown
Break [theory] into first principles. Output: 4 base assumptions, the logical chain from assumptions to theory, and 1 example of what changes if any assumption is wrong.3. 3-analogy layer
Give 3 analogies for [theory] at increasing fidelity. Analogy 1: simplest, most wrong. Analogy 2: more accurate, slightly harder. Analogy 3: closest to reality, hardest. Mark what each analogy gets right and wrong.4. Worked-example simplifier
Apply [theory] to a real problem: [problem]. Output: setup, the theory steps applied, the answer, and 1 sentence on how the theory simplified the problem.5. Anti-example pair
Give 2 examples that look like [theory] applies but actually do not. For each, explain why it fails and which assumption is violated.6. 3-tier explanation
Explain [theory] at 3 levels: (a) headline (1 sentence), (b) intuition (1 paragraph), (c) formal (with notation). Mark what is lost going from formal to intuition.7. Common-misunderstanding list
List the top 5 misunderstandings of [theory]. For each: the misconception, why it spreads, the correction in 1 sentence.8. Visual / diagram description
Describe a diagram I could draw to teach [theory]. Output: what to draw, what each element represents, and the 1 question to ask after showing the diagram.9. Real-world consequence
What changes in the real world if [theory] is true vs false? Give 3 observable differences and 1 experiment that could distinguish them.10. Historical-context simplifier
Explain [theory] by walking through the problem it solved when introduced. Output: the world before, what was missing, how the theory fixed it, what new questions it opened.11. Adjacent-theory bridge
I already understand [known theory]. Explain [new theory] as an extension or contrast to [known theory]. Mark which parts are shared, which are new, and what new assumption [new theory] adds.12. “What you should not take from the simplified version” appendix
Below is my simplified explanation of [theory]. List the 5 things a reader could mistakenly take away from this simplification. Add 1 line each correcting it.
[paste simplified explanation]Common mistakes
- Simplifying so far the theory is technically wrong — clean explanation, confidently false. Always run prompt #12 afterward.
- No worked example — the reader thinks they get it until they try to apply it. Prompt #4 fixes this.
- No anti-example showing where the theory does NOT apply, which leads to over-extension. That is prompt #5.
- Skipping the historical reason it was invented, which strips meaning from the formalism (prompt #10).
- No “what is still hard about this” honesty — it sets up false confidence for the next, harder topic.
- One analogy doing all the work. Every analogy breaks somewhere; prompt #3’s tiered approach makes the breaks visible.
FAQ
Why not just ask the model to “explain it simply”? A single “explain simply” request optimizes for sounding clear, not for staying true. The model drops caveats to read smoothly, and you can’t see what it cut. The prompts here split the job: one part simplifies, another part (prompts 5, 7, 12) names what the simplification got wrong.
Which AI is best for this in mid-2026? For honest, step-by-step explanations, Claude (Opus 4.7 or Sonnet 4.6) is the strongest because it surfaces its own uncertainty. ChatGPT GPT-5.5 in Thinking mode is close and slightly better at tidy worked examples. Use Gemini 3.1 Pro when you want current real-world data folded into the examples, since it searches live.
Do these work on the free tiers? Yes — all three free tiers run the explainers. The limit you’ll hit is session length: ChatGPT Free and Claude Free throttle long back-and-forths, so for a deep multi-prompt session a $20/mo plan (Claude Pro, ChatGPT Plus, or Google AI Pro at $19.99) is worth it.
How do I verify the simplification is actually correct? Run prompt #12 last, pasting back the explanation you received. Then sanity-check the corrections against one authoritative source (a textbook chapter or Stanford Encyclopedia of Philosophy entry for conceptual topics). Never ship an AI explanation without the guard step.
Can I chain these prompts? Yes, and that’s the intended flow: start with #6 (three-tier) for the backbone, add #3 (analogies) and #4 (worked example) for grip, then close with #5 (anti-examples) and #12 (misinformation guard). The Feynman explainer (#1) works as a fast standalone when you just need the gist.