Anthropic Prompting Rules Compilation

1. DEFINE THE OUTPUT - NOT THE REQUEST → Don’t say: "help me with this" → Say exactly what you need: thread, email, table, script, JSON → Clear outputs = better responses 2. SET THE LENGTH UPFRONT → "5 bullets." "200 words." "3 examples." → Define structure before the model writes → Add: "No fluff. No intro. No conclusion." 3. USE POSITIVE INSTRUCTIONS → AI responds better to "do this" than "don't do this" → Replace "don't be vague" with "be highly specific" → Guide the result instead of blocking it 4. START WITH STRONG ACTION WORDS → Skip: "Can you help me..." → Use: Write. Build. Analyze. Create. → Direct prompts create sharper outputs 5. MAKE THE MODEL THINK DEEPER → Add: "Think step-by-step before answering." → Better reasoning = better quality → Most users never do this 6. PUSH FOR HIGHER QUALITY → Add: "Go beyond the obvious." → Or: "Make this production-ready." → One sentence can completely upgrade the result 7. TEACH IT YOUR WRITING STYLE → Give 2-3 examples in your tone → Say: "Match this style exactly." → Reuse your best examples forever 8. CONTROL TOOL USAGE → For research: "Use search and cite sources." → For speed: "Answer using training knowledge only." → Great prompting also means controlling tools 9. EXPLAIN THE GOAL FIRST → Start with: "Goal: ..." → Define the audience and outcome clearly → Weak goals create weak outputs