AI glossary Models & Architecture
What is chain-of-thought (CoT) prompting?
Chain-of-Thought (CoT)
← All glossary termsA way of prompting or training a language model to write out intermediate reasoning steps before giving its final answer.
Explained
Chain-of-thought (CoT) is a technique in which a large language model produces intermediate reasoning steps before its answer, instead of answering directly. It was introduced as a prompting method: showing the model a few examples that include worked reasoning improved its performance on arithmetic, commonsense and symbolic reasoning tasks. Today many models are trained to reason step by step on their own. CoT is used in agent planning and tool selection, and the written steps help people review how an answer was reached, although they are not a guaranteed record of the model's internal process.
How chain-of-thought works
In its original form, the prompt includes a few examples where each question is followed by a worked solution and then the answer. The model imitates that pattern and writes its own steps before concluding. A simpler variant just asks the model to reason step by step.
Writing the steps gives the model more room to break a problem down, keep track of intermediate results and catch some mistakes along the way. The gains are largest on tasks that genuinely need several steps; on simple lookups they are small or absent.
Chain-of-thought and reasoning models
Several recent models are trained to produce extended reasoning before answering, sometimes hidden from the user and sometimes summarised. This moves chain-of-thought from a prompting trick to a built-in capability.
For teams building on these models, explicit step-by-step instructions matter less than before. Clear goals, good context and the right tools matter more, as does deciding how much reasoning a task is worth, since longer reasoning costs more tokens and time.
Limits to keep in mind
- The written reasoning can look convincing and still be wrong.
- It is not a faithful window into how the model computed its answer, so it should not be treated as an audit trail on its own.
- More reasoning means more tokens, higher cost and higher latency.
- For decisions that matter, check the result against data or rules, not only the explanation.
Frequently asked questions
What is an example of chain-of-thought prompting?
Instead of asking only for the answer to a word problem, the prompt shows an example where the solution is worked out: first the quantities, then each calculation, then the result. Given a new problem, the model follows the same pattern and writes its steps before answering.
What is zero-shot chain-of-thought?
It is the variant where no worked examples are given; the prompt simply asks the model to think step by step. It is easier to use than the few-shot version, because nobody has to write example solutions.
Does chain-of-thought reduce hallucinations?
It can reduce reasoning errors on multi-step problems, but it does not stop a model from inventing facts. A model can reason carefully from a wrong premise. Grounding answers in retrieved sources and verifying outputs remain necessary.
Is chain-of-thought still useful with reasoning models?
Less as a prompting technique, because these models already reason before answering. The idea remains useful when designing agents and workflows: breaking a task into explicit steps that can be checked is still good practice.
Can chain-of-thought be used to explain AI decisions?
Only partly. The steps help a reviewer follow the logic, but they are generated text, not a reliable trace of the model's internal computation. For regulated decisions, combine them with logs, rules and other explainability methods.
Decision guides that use this term
Where it shows up at Thinkia
Further reading
Related terms
Sources
- Wei et al. · Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Kojima et al. · Large language models are zero-shot reasoners (arXiv)
- Anthropic · Reasoning models don't always say what they think
Updated: 2026-10-21