Most bad AI output isn't the model being dumb. It's the prompt being vague. "Write a product description" gives the model a thousand reasonable directions and no way to pick yours — so it picks the blandest average of all of them. A sharp prompt removes the guesswork.
Here are the seven moves that do most of the work. You don't need all of them every time — but every one you add narrows the model toward your answer instead of the average one.
The seven moves
- Role — tell it who it's being. "You are an expert copywriter" isn't magic, but it shifts the model toward the right register and vocabulary.
- Audience — who's reading? A line for beginners and a line for experts are different texts. Name the reader.
- Context — give it the actual facts, and tell it not to invent beyond them. This single instruction kills most hallucination.
- Constraints — length, tone, what to avoid. "Under 120 words, plain language, no clichés" does more than any amount of "make it good."
- Format — the shape you want back: bullets, a table, JSON, three options. Ask for the format you'll actually use, or you'll reformat it by hand.
- Examples — paste one or two outputs you like. The model matches patterns better than it follows adjectives. This is the highest-leverage move of the seven.
- Think first — for anything reasoning-heavy, "think step by step before the final answer" trades a little length for a lot of accuracy.
Build one yourself
Edit the task, then click the techniques on and off. Watch the prompt assemble in real time — and feel how much sharper the "everything on" version is than the bare task you started with.
Copy whatever you build straight into your own chat model. Then change the task to your real job and do it again — the muscle you're building is adding the missing constraints automatically, which is most of what good prompting actually is.