The practice of phrasing your input to get better results. The reliable basics: be specific about what you want, state the format you need, give an example or two, and provide relevant background rather than assuming it is known. Asking a model to work through a problem step by step often produces noticeably better answers on anything requiring reasoning.
Technical view
Established techniques include few-shot prompting, supplying worked examples in the prompt; chain-of-thought prompting, eliciting intermediate reasoning steps before an answer; and structured output constraints. Newer reasoning-focused models do much of this internally and can respond poorly to prompting styles designed for earlier generations. The field moves quickly and specific tricks age fast, while the general principle of clear, specific, well-contextualised instruction does not.
Related
Part of AI Terminology Index, section: Building and Using Models.