Chain of ThoughtCoT

Chain of Thought (CoT) prompting instructs a model to break its reasoning into explicit steps before producing a final answer. This dramatically improves accuracy on multi-step tasks like math, logic puzzles, and complex planning.

How it works: Simply adding "Let's think step by step" or "Explain your reasoning before answering" to a prompt causes most modern LLMs to generate intermediate reasoning steps. These steps then inform the final answer, reducing errors that arise from trying to jump directly to a conclusion.

Variants:

• Zero-shot CoT — just append "Think step by step" to any prompt
• Few-shot CoT — provide worked examples showing the reasoning process
• Self-consistency — sample multiple CoT chains, take the majority answer
• Tree of Thought — explore multiple reasoning branches simultaneously
• Extended thinking — Claude and some models support a dedicated reasoning phase

When to use it: Anytime the task requires more than one logical step — arithmetic, multi-condition decisions, code planning, scientific reasoning. For simple factual recall, CoT adds tokens without benefit.

The tradeoff: More tokens = more cost and latency, but much higher accuracy on hard problems.