The problem with most prompts
Most disappointing AI output traces back to one thing: the prompt described a topic, not a task. "Tell me about pricing strategy" gives the model almost nothing to aim at, so it hands you the average of everything it knows about pricing strategy. Broad and forgettable.
A task has an actor, an action, and a shape. "You are a pricing consultant. Write three pricing tiers for a solo web design business, with a one-line reason a client would pick each one" gives the model something to build toward.
Build the prompt in three parts
Who is answering. Give the model a role when the role changes how it should answer. "As a copy editor" produces different output than "as a growth marketer" for the same sentence.
What they are doing. Name the actual task: writing, summarizing, comparing, critiquing, converting. Not "help me with this email," but "rewrite this email to be shorter and more direct."
What done looks like. State the output shape up front: a table, three bullet points, a single paragraph under 100 words. Without this, the model guesses, and it usually guesses long.
Try it
Take a prompt you have used before and check it against those three parts. If any one is missing, that is the gap.
Before: "Give me ideas for a landing page."
After: "You are a conversion copywriter. Write three headline options for a landing page selling a monthly web maintenance plan to small business owners. Each headline under 10 words, one line explaining who it targets."
The second version will not need a follow-up prompt to fix it. That is the real test of a good prompt: does the first answer actually work.