Chain-of-Thought Prompting
Last updated June 11, 2026
What is Chain-of-Thought Prompting in simple terms?
In simple terms, chain-of-thought prompting is telling an AI to show its working instead of jumping to an answer. Like a student who writes out each step of a maths problem, it makes fewer mistakes.
What is Chain-of-Thought Prompting?
Chain-of-thought prompting is a technique that asks a language model to work through its reasoning step by step before giving a final answer, which improves accuracy on problems that need several stages of thinking.
Ask a language model a question that needs several steps of thought — a word problem, a logic puzzle, a question with a few twists — and if it tries to leap straight to the answer, it often trips up, because it's effectively guessing the conclusion in one go. Chain-of-thought prompting is a simple but powerful fix: instead of asking only for the answer, you prompt the model to lay out its reasoning step by step first, and the answer at the end. Often this is as easy as adding a phrase like "let's think step by step" to your request. By spelling out the intermediate steps, the model gives itself room to work the problem through, and the final answer it lands on is far more likely to be right.
Why does merely asking for the steps help so much? A language model produces its output one piece at a time, each piece influenced by what came before. When it writes out a reasoning chain, each step becomes part of the context the next step builds on, so the model is essentially scaffolding its own thinking — breaking a hard leap into a series of small, manageable ones. It's the same reason a person rarely solves a tricky calculation purely in their head but gets it right on paper: writing the steps down both reduces the load at any one moment and exposes a wrong turn so it can be caught. Chain-of-thought is the prompting technique; it's also the idea that newer reasoning models bake in automatically, doing this step-by-step working internally without being asked.
It's a staple of prompt engineering precisely because it's cheap and effective — a few extra words can meaningfully lift performance on anything involving reasoning. But it isn't a cure-all. A model can produce a tidy, plausible-looking chain of reasoning that nonetheless contains a flawed step and arrives at the wrong answer, so a confident chain of thought is not proof of a correct one. And on simple, direct questions it just adds length for no benefit. Used where it fits — multi-step problems — it's one of the highest-value tricks for getting better answers out of an AI.
Real-world example of Chain-of-Thought Prompting
Suppose you ask an AI: "A shop sells pens in packs of 6. I need 40 pens — how many packs, and how many spare pens will I have?" Asked for just the answer, a model might blurt "6 packs" and stop, quietly wrong. Prompted to think step by step, it works it through: 40 divided by 6 is 6 with a remainder, 6 packs give 36 pens, that's short of 40, so you need a 7th pack for 42 pens, leaving 2 spare. Spelling out each step is what catches that you must round up — and turns a confident wrong answer into a correct one. The only change was asking it to show its working.
Related terms
Frequently asked questions about Chain-of-Thought Prompting
What is the difference between chain-of-thought prompting and a reasoning model?
Chain-of-thought prompting is a technique you apply — you word your prompt to ask any model to reason step by step. A reasoning model is a model trained to do that step-by-step working automatically, internally, whether or not you ask. In effect, reasoning models have chain-of-thought built in and taken further. With an ordinary model you have to prompt for the steps; with a reasoning model the careful working happens on its own, which is why reasoning models do better on hard problems out of the box.
How does chain-of-thought prompting work?
You prompt the model to write out its reasoning before the final answer — often just by adding "let's think step by step." Because a model generates text one piece at a time, with each piece shaped by what came before, the written-out steps become scaffolding the next step builds on. This breaks one hard leap into several small ones, lowering the chance of error at each point and making wrong turns easier to catch — much like solving a calculation on paper instead of entirely in your head.
What is chain-of-thought prompting used for?
It's used to boost accuracy on anything that needs multi-step thinking: arithmetic and word problems, logic and reasoning puzzles, multi-part questions, and tasks where the model must combine several facts to reach a conclusion. It's a core prompt-engineering technique because it's cheap — a few words — and effective. It adds no value on simple, direct questions, where it only lengthens the answer, and it's no guarantee of correctness, since a neat-looking chain can still contain a flawed step.