Prompt Engineering / Chain-of-Thought
Rewrite any prompt to require explicit reasoning before answering — preventing intuitive leaps.
Why This Prompt Exists
Models guess. When they guess correctly, you trust them. When they guess wrong, you’re confused why. Forcing reasoning prevents guessing.
You get:
- correct answers with no insight into how the model got there
- wrong answers that seem plausible (because you can’t see the flawed reasoning)
- inability to debug or improve the prompt (no trace to examine)
- overconfidence in model outputs (they sound confident even when wrong)
- repeated failures on the same type of problem (no learning from mistakes)
But explicit reasoning changes everything:
- forces step-by-step: model can’t jump to conclusion without walking through it
- reveals assumptions: what the model thinks is true (but may not be)
- enables debugging: you can see where the reasoning goes wrong
- improves accuracy: models with reasoning steps are more accurate
- builds trust: you can verify the logic, not just the answer
Without forcing reasoning, you accept guesses.
This prompt rewrites any task to require explicit step-by-step reasoning.
The Prompt
Assume the role of a reasoning engineer who forces explicit step-by-step thinking. Your task is to rewrite a prompt to require reasoning before the answer. Generate: 1. ORIGINAL PROMPT - The prompt as written (likely missing reasoning instructions) 2. REASONING REQUIREMENTS - What steps are needed to solve this problem? - What assumptions must be stated explicitly? - What edge cases need checking? 3. REWRITTEN PROMPT (with forced reasoning) - Add instruction: "Before giving your final answer, show your reasoning step by step." - Structure: "Step 1: [first step]. Step 2: [second step]. ... Final answer: [answer]." - Require explicit assumption statements - Require verification of each step 4. BEFORE/AFTER COMPARISON - Show how the rewritten prompt changes model behavior 5. WHEN TO USE (and when NOT to use) - Use for: math, logic, diagnosis, planning - Avoid for: creative tasks, summarization, tasks where speed > accuracy 6. READY-TO-USE PROMPT - Copy-paste version of the rewritten prompt INPUTS: Original prompt: [PASTE THE PROMPT THAT NEEDS REASONING] Task type: [MATH / LOGIC / DIAGNOSIS / PLANNING / OTHER] Desired detail level: [MINIMAL (just key steps) / STANDARD / VERBOSE (explain every assumption)] Model: [GPT-4 / CLAUDE / GEMINI] RULES: - Always separate reasoning from final answer (use headings or delimiters) - Require explicit restatement of the problem before solving - Require assumption checking (what am I assuming that might be false?) - Require step verification (how do I know this step is correct?) - For multi-path problems, require consideration of alternatives
How To Use It
- Run this on any prompt where accuracy matters more than speed.
- Use the rewritten prompt for math, logic, and diagnostic tasks.
- Don’t force reasoning for creative tasks (it can stifle creativity).
- Train your team to recognize when a prompt needs reasoning vs. when it doesn’t.
- Save the rewritten prompt as a template for similar tasks.
Example Input
Original prompt:
“What is 15% of 280?”
Task type:
“MATH”
Desired detail level:
“STANDARD”
Model:
“GPT-4”
Why It Works
Most prompts ask for answers directly — which works for simple recall but fails for reasoning tasks.
This framework improves outcomes by forcing:
- step-by-step structure (no leaps allowed)
- explicit assumptions (what the model takes for granted)
- verification requirements (checking each step)
- separation of reasoning from answer (auditable trace)
- edge case consideration (what could go wrong)
Great step-by-step forcing doesn’t slow down the model meaningfully — it prevents wrong answers.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.