Using an AI tool to help with writing works the same way delegating to a good assistant would: the quality of what comes back depends almost entirely on how clearly you asked. A vague prompt produces generic prose because generic is the only thing a vague request can specify.
Give it a job description, not a topic
Weak prompt: "Write about time management." This has no audience, no angle, and no constraint, so it returns the most average version of that topic imaginable. Stronger prompt: "Write for a freelancer who feels busy but unproductive. Focus on the difference between urgent and important tasks. Keep it under 300 words and use one concrete example." You've now specified the reader, the angle, the length, and the proof — there's far less room left for filler.
Ask for a draft to edit, not a finished piece
Treat the output as a first pass, not a final one. It's often easier to spot what's wrong with existing text than to generate ideas from nothing, so let the tool produce a rough version and then do the real work: cutting the generic phrases, adding a specific example only you would know, and reading it aloud to catch anything that sounds like nobody actually talking.
- Specify tone with a comparison, not an adjective — "like a colleague explaining this over coffee" beats "friendly tone"
- Ask for the weak version to be flagged, e.g. "tell me which sentences are generic"
- Iterate in small steps rather than asking for a full rewrite each time
Keep the judgement yours
The prompt gets you a draft; it doesn't get you the decision about what's true, what's worth saying, or what your reader actually needs. That part stays with you, and it's the part that makes the difference between a passable draft and a piece worth publishing.
