2.4 — Rule #3 — Make the AI think out loud (the "Chain-of-Thought")

🟢 In plain words — instead of asking for only the answer, you ask the AI to show its reasoning first. Like a student writing out the calculation: spelling out the steps reduces careless mistakes.

Pose a math or logic problem and demand the answer directly: a standard model may get it wrong. The solution is called Chain-of-Thought (CoT), or “chain of thought.”

You can see these three techniques as a rise in sophistication: from a prompt with no example (zero-shot) to a prompt with examples (few-shot), then to a prompt that asks for spelled-out reasoning (Chain-of-Thought).

Left to right: you guide the model more and more.

• The trigger: add “Let’s think step by step,” or explicitly ask it to detail its reasoning before the conclusion.
• Why it works: instead of blurting out an answer, the AI lays out its reasoning. Like at school: writing out the calculation avoids careless mistakes.

Caution

⚠️ 2025-2026 update — must know. “Let’s think step by step” was a revolution… for standard models. But the recent reasoning models (models that think internally before answering — Claude with extended thinking, ChatGPT’s Thinking/Pro modes…) already reason internally: explicitly asking them to think step by step is often redundant, sometimes counterproductive. Practical rule:

• Standard / fast model → CoT helps a lot. Use it.
• Reasoning model → let it think; focus your prompt on a clear goal and good criteria, not on “think step by step.”

✅ In short

• Chain-of-Thought makes the AI detail its reasoning before the answer.
• Very useful on standard/fast models (math, logic).
• On recent reasoning models, it’s often redundant: aim for a clear goal and criteria instead.
• CoT (reasoning) and Few-shot (format) combine without trouble.