Prompting
Chain-of-Thought Prompting
A prompting technique that asks the model to reason step-by-step before producing a final answer, often improving accuracy on multi-step tasks.
Definition
Chain-of-thought prompting instructs a language model to show its reasoning explicitly — listing intermediate steps, considering alternatives, then arriving at a conclusion — rather than jumping straight to a final answer. The pattern was first formalised in research showing that for tasks requiring multi-step inference (arithmetic, logic, complex extraction), this kind of structured prompting reliably improves accuracy, especially in larger models.
Why it matters
For writing work, chain-of-thought matters in two situations. First, when a rewrite involves a decision — does this preserve the original meaning, does this match the requested tone, does this stay within length — having the model articulate its check before producing output catches errors. Second, when output must satisfy multiple constraints, an explicit reasoning step makes the final answer more reliable. The trade-off is verbosity. Visible reasoning consumes tokens (and therefore cost and latency). Many production systems use a 'silent' variant: the model is instructed to reason internally and return only the conclusion.
Examples
Direct prompt
Rewrite this for a senior executive. Draft: [...]
Chain-of-thought prompt
Identify the three changes needed to make this draft suitable for a senior executive, then apply them. Draft: [...]
Frequently asked
Does chain-of-thought work on every task?
It helps most on multi-step reasoning. For simple instructions, it adds cost without improving quality.
Should the user see the reasoning?
Usually not in a finished product. Reason internally, present a clean answer.
Does it work on small models?
Less reliably. The technique was discovered as an emergent capability of larger models.
Related terms
Prompt Engineering
The discipline of crafting model inputs — instructions, context, examples, and constraints — to produce reliably better outputs.
Few-Shot Prompting
Showing a model two to five worked examples of the desired input-output pattern before asking it to handle a new case.
System Prompt
A persistent set of instructions that frames a model's behaviour across an entire interaction or product.
Put the concept to work
Open the rewrite engine and apply this principle to a draft of your own.
Try a rewrite