If AI output keeps coming back generic, the prompt is usually the problem. These four adjustments do most of the work.
Say who it is for
"Write a product description" and "Write a product description for a first-time buyer who is not sure they need this" produce very different results. The audience shapes vocabulary, length and which objections get addressed. State it explicitly.
Give it an example of what good looks like
Describing a tone rarely works as well as showing one. Paste in a paragraph you like and say "match this voice." The model is far better at imitating a concrete sample than interpreting adjectives like "professional but friendly," which could mean almost anything.
Constrain the shape of the answer
Say how long, what structure, and what to leave out:
- "Three bullet points, under twelve words each."
- "No introduction, start with the first step."
- "Plain sentences. No marketing adjectives."
Without constraints you get a default shape - an intro, a list, a summary - whether or not it suits your use.
Iterate instead of restarting
The first output is a draft to react to. "Cut the first paragraph and make the last point more specific" gets you further than writing a new prompt from scratch. You are editing, and editing is faster than commissioning.
Keep the ones that work
When a prompt produces something genuinely good, save it with a note on what it was for. Within a few weeks you will have a small personal library that beats any generic prompt pack, because it is tuned to your actual work.
The pattern underneath all four: the model cannot read your context. Everything you have not said, it will guess at, and the average guess produces average writing.
