What is chain-of-thought prompting?

Chain-of-thought (CoT) prompting asks the model to work through a problem step by step before giving a final answer, rather than jumping straight to a conclusion. Producing the intermediate reasoning gives the model more 'room' to compute, which measurably improves accuracy on multi-step arithmetic, logic, and commonsense tasks.

What is zero-shot chain-of-thought?

Zero-shot CoT is the discovery that simply appending a phrase like 'Let's think step by step' to a prompt triggers step-by-step reasoning without needing any worked examples. It is the simplest form of CoT — one line — and often gives a large accuracy boost on reasoning tasks for almost no effort.

Does chain-of-thought help on every task?

No. CoT helps most on problems that genuinely require multiple reasoning steps — math word problems, logic puzzles, multi-hop questions. For simple lookups or short factual answers it adds length and cost without benefit, and can even introduce errors by over-reasoning. Use it where the task has real intermediate steps.

What is least-to-most prompting?

Least-to-most prompting first asks the model to break a hard problem into a sequence of simpler subproblems, then solves them in order, feeding each answer into the next. It pushes chain-of-thought further by making the decomposition explicit, which helps on problems that are too complex to reason through in one pass.

Chain-of-Thought Prompting: Getting AI to Show Its Work

This page explains why making a model reason step by step improves its answers, and gives you a builder that wraps your own question in a proven chain-of-thought prompt you can copy into any LLM.

Why reasoning aloud helps

Chain-of-thought (CoT) prompting was introduced in research showing that when a large model is prompted to produce intermediate reasoning steps before its final answer, accuracy on multi-step problems rises sharply. The intuition is that a model generates one token at a time, so forcing it to lay out the steps gives it more computation to reach the answer, instead of trying to leap to a conclusion in a single step. The effect is strongest on arithmetic word problems, logic, and multi-hop questions — and it grows with model size.

Zero-shot CoT and structured variants

The most striking follow-up finding is zero-shot CoT: simply adding “Let’s think step by step” to a prompt triggers the same step-by-step behaviour without any worked examples. It is a single line that often delivers a large accuracy gain. You can make this more reliable by asking for structured steps — numbered reasoning, or a clear separation between the working and the final answer — which also makes the output easier to check. The interactive builder below produces both the simple zero-shot version and a structured version from your question.

Least-to-most prompting

For problems too hard to reason through in one pass, least-to-most prompting decomposes the task first. The model is asked to break the problem into a sequence of simpler subproblems, then to solve them in order, carrying each result forward. This explicit decomposition helps on compositional problems where naive CoT still stumbles, because it stops the model from biting off too much at once.

When to use it — and when not to

Use chain-of-thought where the task has genuine intermediate steps: math, logic, planning, multi-part questions. Avoid it for simple factual lookups or short answers, where the extra reasoning adds cost and latency and can even introduce mistakes through over-thinking. A practical rule: if you would naturally work the problem out on paper, CoT will probably help; if the answer is a single fact you would just recall, it will not.