Write concise, well-specified prompts to reduce token usage and the chance of needing a retry.
Easy to implement
5 High impact score
A vague or overlong prompt costs more tokens outright, and it also raises the odds the model needs a follow-up or clarification round to land on a usable answer; multiplying the token cost of the whole interaction rather than just the first message. Be explicit about intent, include only the context that’s actually relevant to the task, and specify the output format up front rather than leaving it to guesswork. Prompt engineering is a sustainability lever as much as it is a quality one.