There’s a lot written about prompting and most of it is decoration. Strip it back and almost every good prompt has the same four parts. Get those in and the model’s first answer is usually the one you wanted.
1. Goal
One line. What you want back. “Build a landing page for a macOS dictation app.” “Rewrite this email so it’s shorter and less apologetic.” If you can’t say the goal in a sentence you don’t know what you want yet, and no prompt will fix that.
2. Context
What the thing is for, who it’s for, and what already exists. This is the part people skip when typing because it’s tedious, and it’s the part the model needs most. “The audience is Mac users who write all day.” “This goes to a customer who’s already annoyed.” “We use Postgres and the schema is attached.”
3. Constraints
The “no”, the “must” and the “keep it”. “No Tailwind.” “Keep the function signature.” “Under 150 words.” “British spelling.” Constraints are cheap to state and expensive to leave out, because the model will happily produce something perfect that violates one.
4. Output
The shape of what comes back. A file. A table. Three options with a recommendation. A paragraph, not bullets. Models do what you ask; the failure is usually that you didn’t ask.
The template
Say it, don’t type it:
Goal: [one line]. Context: [two sentences]. Constraints: [the nos and musts]. Output: [the shape].
Or just ramble with all four in there somewhere. Speaky Box’s Smart key pulls them out of what you said and lays them out in this order, which is the whole reason it exists. Either way, four parts, every time, and you’ll stop having the same conversation three times.