What actually makes a prompt good, if longer is not better?
Making my prompts longer has not made the answers better. For people who prompt every day, which single thing made the biggest difference: more context, a role, examples, a clear structure, a stated output format, or something else?
Clarity beats length. The prompts that work tend to answer four things in plain words: what the situation is, what you need, what a useful answer looks like, and what to avoid. Extra paragraphs that do not change any of those four only give the model more to misread.
If you add one thing, add an example. One good input and output pair usually does more for style and format than another 200 words describing them. Pick examples that cover the tricky case, not only the easy one.
State the output format when you will reuse the answer: a table, a list of a fixed length, or JSON with named fields. A role line is the weakest lever of the set; it can set tone, but it does not add knowledge the model lacks.
Then test, do not guess. Keep the prompt, run it on a few real inputs, change one thing at a time and keep the version that wins. That habit is what separates a prompt that worked once from one that works every time.
Listings mentioned
- Prompt engineering patterns · skill by wshobsonCovers few-shot examples, structured output and step-by-step reasoning, with when each one is worth using.
- Meta prompt engineer · prompt by magejoshA reusable prompt that asks you clarifying questions first, then writes the prompt for your goal.
Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.