Chain-of-Thought Prompting: How to Get AI to Reason Step by Step
By Emmanuel Abou Chabke, Founder and Editor · Reviewed by the AIQuickPrompt editorial team · About the publisher
Chain-of-thought prompting means asking an AI model to work through a problem in steps before it gives the final answer. It was first described in a 2022 research paper, and it still helps on multi-step tasks like maths, planning, comparisons and debugging. This guide explains how it works, when it is worth the extra tokens, how it differs from prompt chaining, and gives you templates you can save in AIQuickPrompt and reuse.
What is chain-of-thought prompting?
Chain-of-thought (CoT) prompting is a technique where you ask the model to break a problem into intermediate steps before answering. Instead of jumping straight to a result, the model writes out the reasoning, which tends to reduce mistakes on tasks that need several steps.
The idea comes from the paper "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models" by Wei and colleagues at Google (2022). They showed that a few worked examples with written-out reasoning improved results on arithmetic, commonsense and symbolic reasoning tasks for large models.
Zero-shot vs few-shot chain-of-thought
There are two common ways to use it. Zero-shot CoT simply adds an instruction such as "Think through this step by step, then give the final answer." Few-shot CoT includes one or more worked examples that show the reasoning pattern you want, then asks the new question.
Zero-shot is quick and costs fewer tokens. Few-shot gives you more control over the format and depth of the steps, and pairs naturally with the techniques in our few-shot prompting guide.
- Zero-shot CoT: one extra instruction, good default for everyday questions.
- Few-shot CoT: worked examples, best when you need a consistent structure.
- Structured CoT: ask for reasoning and final answer in separate, labelled sections so the answer is easy to find.
When does chain-of-thought help, and when is it unnecessary?
CoT is most useful for problems with several dependent steps: calculations, comparing options against criteria, planning a schedule, troubleshooting code or checking a document against rules. For simple lookups, rewrites or short creative tasks it mostly adds length.
Many current models, including the reasoning modes in recent GPT, Claude and Gemini releases, already think internally before answering. Provider guidance notes that these models often do best with clear, direct instructions rather than detailed step-by-step scripts. With them, ask for a short summary of the key steps if you want to check the logic, rather than forcing long visible reasoning.
- Use it: multi-step maths, decisions with trade-offs, debugging, rule checking.
- Skip it: translation, tone rewrites, short summaries, simple facts.
- Test it: run the same prompt with and without the instruction and compare.
Chain-of-thought vs prompt chaining
The names are similar but the techniques are different. Chain-of-thought happens inside one prompt: the model reasons step by step in a single answer. Prompt chaining splits a task across several separate prompts, and you pass the checked result of one into the next.
They work well together. A single step in a chain can use CoT when that step needs careful reasoning. Read our prompt chaining guide for the multi-prompt version.
Reusable chain-of-thought templates
Save these as prompts in AIQuickPrompt and fill the {{variables}} each time with Use Prompt.
Decision template: "I need to choose between {{options}} for {{goal}}. My criteria are {{criteria}}. Work through each option against each criterion step by step, using only the facts I gave you. If a fact is missing, say so instead of guessing. Then give a final recommendation in one sentence under the heading Answer."
Calculation template: "Solve this: {{problem}}. Show each step of the calculation on its own line, check the result by a second method, then give the final number under the heading Answer."
Debugging template: "Here is my code: {{code}}. The error is: {{error}}. Reason step by step about what could cause it, list the most likely cause first, then give the smallest fix."
Does chain-of-thought cost more AI credits?
Yes, usually. Written reasoning means more output tokens, and output tokens are what most providers charge for. On a cheap fast model that is often still less than running a premium model without CoT, but it depends on the task.
A practical approach: try a fast model with a CoT instruction first, and only move to a premium model if the answer is still wrong. Our guide to saving AI credits covers more ways to keep costs down.
How AIQuickPrompt helps you use chain-of-thought
Store your best CoT templates in a folder so they are one tap away in any AI tool. Variables turn each template into a fill-in form, and version history lets you compare a version with step-by-step instructions against one without.
AI Optimise can rewrite a prompt for you, and Guide the optimisation lets you ask it to "add step-by-step reasoning and a separate final answer". Recommend a model suggests whether a task needs a premium reasoning model or whether a cheaper one will do.
Common chain-of-thought mistakes
Watch out for these:
- Trusting the steps blindly: written reasoning can still contain errors, so check the facts and the final answer.
- Using it everywhere: simple tasks get longer and more expensive without getting better.
- Over-scripting reasoning models: they often perform better with a clear goal than a forced step list.
- No clear final answer: always ask for the answer in a labelled section.
- Letting the model invent missing facts: tell it to say what is missing instead.