Most “explain this” prompts produce middle-school-depth answers that bore advanced readers and confuse beginners. Two things have to line up: the depth has to match the reader, and the angle has to match what’s actually confusing. A definition won’t fix a missing intuition, and an analogy won’t fix a missing prerequisite. The 12 prompts below hit specific depths and specific failure modes, so you stop getting one flat answer for every question. For the long-form workflow, see how to use AI to explain a difficult concept.
TL;DR: Pick the prompt that matches your reader (ELI5 #1, undergrad #2, PhD #3) or the angle that fixes the gap (analogy #5, comparison #6, history #7, misconceptions #10). Replace every {concept} placeholder before sending. For dense or contested topics, turn on a reasoning mode (ChatGPT “Thinking”, Claude Opus 4.7, or Gemini 3.1 Pro) — as of June 2026 OpenAI reports reasoning modes cut major factual errors by up to ~78% versus single-pass answers, which matters when a wrong explanation quietly anchors a wrong mental model.
Which model for which depth (June 2026)
All three frontier chat apps handle these prompts, but they have different strengths:
| Job | Strong pick | Why |
|---|---|---|
| Clear ELI5 / analogy prose (#1, #5, #11) | Claude Sonnet 4.6 / Opus 4.7 | Cleanest natural-language explanations, fewer filler hedges |
| PhD-level + recent advances (#3, #7) | Gemini 3.1 Pro or ChatGPT (Thinking) | Strong reasoning; can browse to ground “last 3 years” claims |
| Spot-the-flaw / two-expert dialogue (#8, #9) | ChatGPT GPT-5.5 (Thinking) | Holds multi-turn structure and a deliberate planted error well |
| Long source pasted in (a whole paper) | Any — all run 1M-token context | Opus 4.7, Sonnet 4.6, and Gemini 3.1 Pro are 1M standard; ChatGPT Plus app context is ~320 pages |
Free tiers (ChatGPT Free on GPT-5.5, Claude Free on Sonnet 4.6, Google AI Free on Gemini) run every prompt here; reasoning modes and the 1M in-app context sit on the paid tiers ($20/mo Plus, $20/mo Claude Pro, $19.99/mo Google AI Pro).
Best for
- Learning a new technical field
- Cross-discipline study (CS reading bio, lawyer reading ML)
- Teaching prep and lecture writing
- Demystifying jargon in a paper or doc
- Onboarding into a new codebase or domain
1. ELI5 (truly)
Explain {concept} to a smart 8-year-old. Use exactly 1 analogy from their world (school, playground, video games). ≤120 words. Avoid all jargon, even "easy" jargon like "algorithm" or "function". End with one sentence the kid could repeat to a friend.2. ELI undergrad (foundational)
Explain {concept} to an undergrad who has foundation in {prereq field} but no exposure to {concept}. Build from what they already know. State which prerequisite each step rests on. ≤250 words. Include 1 minimal worked example.3. ELI PhD (peer-level)
Explain {concept} at a graduate / research level. Skip basic definitions. Focus on: (1) where the field currently disagrees, (2) recent advances in the last 3 years, (3) the open questions a new researcher could attack. ≤300 words. Cite named approaches, not vague "researchers think".4. 3-depth ladder
Explain {concept} three ways: (1) one sentence the layperson gets, (2) one paragraph an undergrad gets, (3) one page an expert finds useful. Each layer builds on the previous without repeating it. Label each layer's intended reader.5. Analogy + caveats
Explain {concept} using a single strong analogy. Then state 3 places where the analogy breaks down and what reality does instead. Reader: smart adult with no field background. Goal: they leave with intuition that won't mislead them later.6. Compare to a similar concept
Explain {concept-A} by comparing it to {concept-B} (which I already understand). Format: (1) where they're alike, (2) where they differ, (3) the diagnostic question that tells you which one applies in a given situation. ≤300 words.7. Historical trace
Explain {concept} by tracing how it evolved: (1) the original problem people were trying to solve, (2) the first attempts and why they failed, (3) the key breakthrough and what it changed, (4) the current understanding. ≤400 words. Name the people or papers where they matter.8. Spot-the-flaw quiz
Explain {concept} in a 200-word block that contains exactly 1 deliberate flaw or oversimplification. End with: "Where is this explanation wrong or misleading?" Wait for my answer, then reveal the flaw and the correct version.9. Two-expert dialogue
Stage a 6-turn dialogue between two experts who disagree about {concept}. Expert A holds the {mainstream / orthodox} view; Expert B holds a {contrarian / heterodox} view. Each turn ≤40 words. They should clarify the real disagreement, not strawman each other.10. Pre-mortem misconceptions
List the 5 most common misconceptions a newcomer holds about {concept}, ranked by how badly each one breaks downstream understanding. For each: state the misconception in 1 sentence, then state the corrected mental model in 1 sentence.11. Worked example backwards
Explain {concept} by starting from a fully worked numerical or concrete example, then peeling back to the general principle. Reader: someone who learns better from examples than definitions. Show the example first, then label which part of it generalizes.12. Tiered Q&A drill
Generate a 5-question Q&A drill on {concept}, ordered from easiest to hardest. Question 1 = recall. Question 2 = comprehension. Question 3 = application. Question 4 = analysis. Question 5 = edge case / open problem. Include answers in a separate block.Common mistakes
- Defaulting to middle-school depth regardless of reader
- Dumping jargon without grounding it in something the reader already knows
- Skipping analogy when one would do the job
- Using an analogy without flagging where it breaks
- Treating “explain” as “define”: definitions don’t build intuition
- Trusting the “recent advances” claims in #3 and #7 without a source — turn on browsing or ask for citations
FAQ
Why does AI default to middle-school depth? Chat models optimize for the most broadly acceptable answer, which lands around general-reader level. Naming an exact reader (“an undergrad who knows linear algebra but not measure theory”) and a word budget forces the depth up or down. Prompts #1 to #4 do this explicitly.
The spot-the-flaw prompt (#8) reveals the flaw immediately. How do I stop that? On a single message the model often answers and self-grades in one turn. Send the prompt, wait for the 200-word block, then reply with your guess in a separate message before asking for the answer. ChatGPT GPT-5.5 and Claude Opus 4.7 hold this two-turn structure more reliably than instant single-pass replies.
Can these explain a whole research paper, not just a term? Yes. Paste the paper and apply prompt #3 (PhD) or #7 (historical trace). As of June 2026 Claude Opus 4.7, Sonnet 4.6, and Gemini 3.1 Pro all take 1M-token context as standard, so a 40-page PDF fits; ChatGPT Plus holds roughly 320 pages of in-app context, with the full 1M reserved for the $200 Pro tier.
How do I trust the “last 3 years” advances in #3 and #7? Those are the lines most prone to fabrication. Turn on a browsing-capable reasoning mode (ChatGPT Thinking, Gemini 3.1 Pro) and add “cite a paper or named method for each advance.” Then spot-check one citation. See why AI makes up facts and how to verify.
Which single prompt should I start with? For a term you’ve never seen, run #4 (the 3-depth ladder): one sentence, one paragraph, one page. It tells you which depth you actually need, and you re-run just that layer with #1, #2, or #3.