The fastest way to take a vague prompt and turn it into one that runs reliably is to specify three things, in this order: who's writing, who they're writing for, and what shape the answer should take.
The scaffold
You are a [ROLE].
You're writing for [AUDIENCE].
Output as [FORMAT].
[ACTUAL TASK]Each line is doing real work:
- Role sets the model's posture. "Senior backend engineer" produces different defaults than "thoughtful technical writer" or "skeptical product manager."
- Audience sets the assumed-knowledge floor. The same explanation looks completely different for "a CTO with 20 years of experience" vs. "a smart non-technical co-founder."
- Format short-circuits the model's default impulse to write a five-paragraph essay. "Bulleted list with one short sentence per item" or "two-column markdown table" or "single tweet under 240 characters" — pick one and stop arguing.
Order matters
Role first, then audience, then format. Reordering changes the priors. Format-first prompts often produce well-shaped slop because the model fixed the shape before deciding what to put in it. Role-first prompts produce things in voice, which is the part that's hardest to fix after the fact.