Chain of thought means an AI writing out the in-between steps of a problem before it gives the final answer, instead of jumping straight to it. It started as a prompting trick (“think step by step”); many newer AI models now do it on their own.
Remember maths at school, when the teacher said “show your working”? Writing each step down made silly mistakes less likely, and let the teacher see where you went wrong. Chain of thought is the AI showing its working.
In a sentence: “The model kept getting the sum wrong until I asked for chain of thought; once it wrote out each step, it caught its own mistake.”
What the word means
The plain point: a “chain” of small reasoning steps, written out one after another, that leads to the answer.
The term comes from a 2022 research paper. It found that showing an AI a few worked examples with their steps written out helped it on maths and logic questions.1
The paper did this with examples in the prompt. That is few-shot prompting: showing the AI what you want instead of only describing it.
Where the word comes from
The name comes from a paper by Jason Wei and eight co-authors, first posted on 28 January 2022. They called their method “chain of thought prompting”: putting a few worked examples, steps included, in the prompt.2
The shorter trick came a few months later. A paper posted on 24 May 2022 by Takeshi Kojima and colleagues showed that simply adding “Let’s think step by step” before the answer helped, with no worked examples at all.3
A simple example
The plain point: the steps come before the answer, not after.
| Style | “A shop has 15 eggs, sells 8 and gets 12 more. How many now?” |
|---|---|
| Straight answer | Just the number: “19.” Right or wrong, you cannot see why. |
| Chain of thought | Steps first: “15 minus 8 is 7. 7 plus 12 is 19. Answer: 19.” Each step can be checked. |
Why it matters to you
The plain point: on newer models you rarely need to ask for it, and the steps you see are not proof the answer is right.
- Reasoning models already do it. A reasoning model works through the problem before answering. OpenAI says asking these models to think step by step is unnecessary.4
- It can still help when thinking is switched off. Anthropic calls the manual version a fallback for that case.5
- The written steps may not be the real reasons. Anthropic researchers warn that the working an AI shows does not have to match how it actually reached the answer.6 So check the answer itself.
Neat working looks convincing. It is still worth checking the final answer on anything that matters.
Often confused with
- Prompt chaining: splitting a job into several separate prompts, where each one uses the last one’s output. Chain of thought happens inside one answer. See prompt chaining meaning.
- Thinking like a person: the steps are generated text, not a mind at work. AI terms people get wrong explains the gap.
- A guarantee: AI models can still reach a confident wrong answer, even with tidy steps. See hallucination meaning.
Questions people ask
Do I still need to say “think step by step”?
Usually not with reasoning models, which work through the problem on their own. OpenAI says the phrase is unnecessary for them.4 It can still help when a model’s thinking mode is off.5
Can you trust an AI’s chain of thought?
Not completely. The steps are useful for spotting a slip, but Anthropic’s researchers found the working shown does not have to match how the model really reached its answer.6 Check the final answer, not just the tidy steps.
What is zero-shot chain of thought?
It is chain of thought without worked examples: you just ask the AI to think step by step. “Zero-shot” means giving the AI no examples, and the 2022 Kojima paper showed that this simple request helps.3 See zero-shot prompting meaning.
Learn more on this site
- Old tricks on new models: Prompting myths checks “think step by step” against vendor guides.
- By model: Prompting for LLMs: ChatGPT, Claude and Gemini.
- The basics: the AI glossary and AI explained for complete beginners.
- Related: Explainable AI meaning.
- Related: Prompting Techniques Explained.
QUICK CHECK · 3 QUESTIONS
Pick an answer to see if you got it. Nothing is recorded.
1. What is chain of thought?
- An AI linking several separate chats together
- An AI writing out the in-between steps before the final answer
- An AI copying answers from the internet
Show the answer
Answer: An AI writing out the in-between steps before the final answer. Chain of thought means the AI shows its working: it writes out the steps of a problem before giving the answer.
2. Do the steps an AI shows prove its answer is right?
- No, they may not match how it really got there, so check the answer itself
- Yes, neat steps mean the answer is correct
- Yes, but only on maths questions
Show the answer
Answer: No, they may not match how it really got there, so check the answer itself. The page says the working an AI shows does not have to match how it actually reached the answer, so the answer itself still needs checking.
3. How is chain of thought different from prompt chaining?
- They are two names for the same thing
- Chain of thought needs several separate chats
- Chain of thought happens inside one answer; prompt chaining splits a job into separate prompts
Show the answer
Answer: Chain of thought happens inside one answer; prompt chaining splits a job into separate prompts. Prompt chaining uses several prompts, each building on the last; chain of thought happens inside a single answer.
Related terms: AI memory · AI prompt generator · AI search engine · Context engineering · Context rot · Context window
- 1Wei et al., Chain-of-Thought Prompting Elicits Reasoning in Large Language Models, arXiv:2201.11903, 28 Jan 2022, read at source 1 Oct 2026“generating a chain of thought -- a series of intermediate reasoning steps -- significantly improves the ability of large language models to perform complex reasoning.” ↑
- 2Wei, Wang, Schuurmans, Bosma, Ichter, Xia, Chi, Le and Zhou, Chain-of-Thought Prompting Elicits Reasoning in Large Language Models, arXiv:2201.11903, v1 28 Jan 2022, read at source 12 Oct 2026“a simple method called chain of thought prompting, where a few chain of thought demonstrations are provided as exemplars in prompting.” ↑
- 3Kojima, Gu, Reid, Matsuo and Iwasawa, Large Language Models are Zero-Shot Reasoners, arXiv:2205.11916, v1 24 May 2022, read at source 12 Oct 2026“we show that LLMs are decent zero-shot reasoners by simply adding ‘Let’s think step by step’ before each answer.” ↑a“we show that LLMs are decent zero-shot reasoners by simply adding ‘Let’s think step by step’ before each answer.” ↑b
- 4OpenAI, Reasoning best practices, undated guide, read at source 1 Oct 2026“Since these models perform reasoning internally, prompting them to ‘think step by step’ or ‘explain your reasoning’ is unnecessary.” ↑a“Since these models perform reasoning internally, prompting them to ‘think step by step’ or ‘explain your reasoning’ is unnecessary.” ↑b
- 5Anthropic, Prompting best practices, undated documentation, read at source 1 Oct 2026“When thinking is off, you can still encourage step-by-step reasoning by asking Claude to think through the problem.” ↑a“When thinking is off, you can still encourage step-by-step reasoning by asking Claude to think through the problem.” ↑b
- 6Anthropic, Reasoning models don’t always say what they think, 3 Apr 2025, read at source 1 Oct 2026“There’s no specific reason why the reported Chain-of-Thought must accurately reflect the true reasoning process”. ↑a“There’s no specific reason why the reported Chain-of-Thought must accurately reflect the true reasoning process”. ↑b
Wei et al., Chain-of-Thought Prompting Elicits Reasoning in Large Language Models, arXiv:2201.11903, 28 Jan 2022, read at source 1 Oct 2026.
The comparison on this page (showing your working in school maths) and the egg example are this site’s own, not claims from these sources.
Free printable AI cheat sheets, and AI news every week.
Sixteen one-page AI guides and checklists in one free PDF, no sign-up. And if you want to keep up, the newsletter covers what changed in AI each week, in plain English, with the sources.
SUBSCRIBE FREE ↗ GET THE CHEAT SHEET PACK