Type "write me something about X" and you get generic mush. The model is only matching your input. Vague in, vague out.
So I stopped writing the request and started asking the model to build it for me. I have it design the ideal prompt first, hand it a sample of how I actually write, then tell it to run its own brief.
You are an expert at writing [document type] for [your audience or role]. Match the voice in this sample: [paste a paragraph or a past piece you've written, more is better]. Before you respond, design the ideal prompt I should give you to produce [the deliverable] on [TOPIC] for [AUDIENCE]. Then run that prompt.
It works for two reasons. First, it forces your thin one-liner to expand into a real brief, with audience, stakes, and structure, before a word gets drafted. Second, the voice sample is the cheat code. The more of your own writing you paste in, the more it comes back sounding like you instead of a press release.
Feed it real context. Who's reading this, what's actually at stake, and a paragraph or two you've genuinely written. The brief it designs is a starting point you can edit before you run it, and the draft that comes back is still yours to shape.