A strong prompt has five ingredients: a clear role for the model (“Act as a…”), enough context to ground the answer, a specific task, a defined output format (bullets, table, steps), and constraints (length limits, things to avoid). Miss one and the output gets noticeably weaker.

Yes. Under about 20 words, prompts are usually too vague for the model to know what you really want. Over about 400 words, key instructions get buried and costs climb. The sweet spot is typically 50–200 words: enough detail to be precise, short enough to stay focused.

Almost always. A role (“You are a senior data analyst”) activates expert-level patterns in the model — better vocabulary, professional standards, and domain methods. The only time to skip it is for trivial tasks like simple rephrasing.

Fix one failed check at a time, starting with the easiest wins: add a role line, then an output format, then one constraint. Re-run the analysis after each change. Most prompts jump from a D to a B with just those three additions.