Claude Cheat Sheet: Write Prompts That Actually Work
A printable one-page reference for getting sharper answers from Claude, with a prompt ladder and ready-to-use patterns.
In this article
Most people use Claude like a search bar and get search-bar answers. The fix isn't a longer conversation — it's a better first prompt. Print this, tape it by your monitor, and level up.
The Prompt Ladder#
Same task, three levels of specificity. Each level adds structure Claude can act on.
Level 1 — Vague
"Summarize this contract." Result: a generic paragraph. Might miss what you actually care about.
Level 2 — Structured
"Summarize this contract in 5 bullet points." Result: scannable, but still generic — Claude picks what seems important.
Level 3 — Targeted
"Summarize this contract in 5 bullet points. Flag any auto-renewal clauses or unusual liability terms. Note the exact cancellation deadline, quoting the section." Result: a summary built around the decision you need to make, with the risky parts called out and no digging required.
The pattern: format (bullets, table, word count) + focus (what matters) + output shape (quotes, deadlines, structure). More constraints, better answers — this isn't a contradiction, it's how these models work.
Reference Table: Prompt Patterns by Use Case#
| Use case | Pattern | Example |
|---|---|---|
| Long document analysis | Ask for structure + page/section references | "Read this 40-page proposal. Summarize in 6 bullets, and cite the page number for each claim." |
| Accuracy-critical tasks | Force citations, forbid guessing | "Only state facts you can quote directly from the text. If something isn't in the document, say 'not stated' — don't infer." |
| Writing/editing in a voice | Give examples + constraints, not just adjectives | "Rewrite this email in a direct, no-fluff tone like our website copy. Max 100 words. No exclamation points." |
| Self-improvement loop | Ask Claude to critique before finalizing | "Critique your last answer for vague language or unsupported claims, then rewrite it fixing those issues." |
| Comparing options | Force a table, define the columns | "Compare these 3 vendors in a table: cost, contract length, cancellation terms, and one risk each." |
| Avoiding hallucinated numbers | Ask it to show its math | "Calculate the total and show each step, so I can verify it." |
Three Habits That Fix Most Bad Answers#
- Tell it the audience. "Explain this like I'm presenting to our board" beats "explain this" every time.
- Ask it to ask you. "Ask me any clarifying questions before you answer" catches missing context you forgot to mention.
- Make it show receipts. For anything you'll act on — contracts, financials, compliance — require exact quotes or page references. If it can't cite it, don't trust it.
The One-Line Upgrade#
If you only remember one thing: after any answer, ask "What's weak or missing in that response? Fix it." This single follow-up catches more errors than a perfectly crafted first prompt.
Prompting well is a skill, and it compounds — teams that get good at it move faster on everything from client emails to contract review. That's a Culture and Talent capability, not just a tooling one, and it's exactly what we measure in our AI Business Maturity Assessment. If your team is still stuck on Level 1 prompts, that's a quick, fixable gap.
