Asking for an answer directly makes the model guess in one jump; asking it to work through the problem makes each step inform the next. 'Let's think step by step' became famous, but explicit structure — 'first extract the quantities, then compute, then state the result' — works even better.
The caveat is that the written reasoning is a generated artefact, not a window into the model: it can be fluent and wrong. For production use the reasoning often runs hidden, with only the checked result shown to the user.
Related terms
Prompt engineering
The practice of writing model inputs so the output is reliably what you need — structure, examples, constraints and tone rather than hope.
Hallucination
A confident, fluent, entirely invented answer — a citation, a figure or an API that does not exist.
Evaluation (evals)
A repeatable test set that measures whether a change to an AI system made it better or worse, rather than just different.
