Formula for a strong prompt
Working template: ROLE (who the model should be) plus CONTEXT (situation and goal) plus TASK (what to do) plus FORMAT (how to return the answer) plus CONSTRAINTS (volume, tone, what to avoid). The more fully these parts are specified, the more accurate the answer.
Template for texts
“You are an experienced editor. Write [what] for [whom]. Tone: [what]. Volume: [limit]. No introductions, just the gist.” Add an example of the desired result - the model will pick up the style (this is a few-shot).
Template for analysis and data
“Analyze [the data]. Return the result in the form [list/table/JSON] with fields [which]. If there is not enough data, mark it, don’t make it up.” The strict format makes the response ready for use.
Template for ideas and solutions
“Offer [N] options [of what] for [situation]. For each there are pros and cons. Then recommend the best one and explain why.” Step-by-step reasoning improves quality on complex problems.
Main principle
Don't leave the format and details to the model's will. The more clearly you describe what you want and how you want it, the less the model guesses. The quality of the answer depends 80% on the prompt.