Reality Lab · 8 MIN · Use
Sharpen a vague request
A one-line prompt gets a generic, off-target answer. Rewrite it so the model spends its effort on your problem, not the average one.
Someone needs to email their team about a delayed launch. They type “write an email about the launch delay” and get back a polished, corporate-sounding message that names no dates, misjudges the audience, and buries the one thing people actually need to know. It reads well and helps no one.
Your task: Rewrite the request so it states the five things the model otherwise has to guess: goal, context, audience, boundary, and format.
"Write an email about the launch delay."
This box is local to your browser. Nothing you type is sent anywhere or saved after you leave.
Check your own work
There's no grade here — checking against these is the exercise. Compare each point to what you wrote.
- The goal is a concrete outcome, not a topic — e.g. “reassure the team and tell them the new date,” not “about the delay.”“About X” lets the model pick any of a hundred purposes. Stating the outcome points its effort at yours.
- You supplied the context only you know — the real new date, the cause, what’s already been communicated.The model can’t invent your facts. Without them it fills the gaps with plausible-sounding fiction, which is exactly what you don’t want in a message people rely on.
- The audience is named, so tone and detail can fit them.An email to your own team reads differently from one to customers or executives. “Audience” is usually the single line a weak prompt is missing.
- A boundary and a format are set — length, what not to promise, and the shape (subject line + short body).Boundaries stop the model over-promising or rambling; a format spec means you get something you can send, not something you have to reshape.
Reveal a worked approach
A sharper version: “Write a short internal email to my product team. GOAL: tell them the launch moved from Sept 3 to Sept 24 because QA found a data-migration bug, and reassure them the scope is unchanged. AUDIENCE: 12 engineers and designers who already knew a slip was possible. BOUNDARY: don’t assign blame, don’t promise the new date is final, keep it under 150 words. FORMAT: a subject line and 2–3 short paragraphs.”
Notice what each line bought you: the goal aimed the message, the context gave it true facts, the audience set the tone, and the boundary and format made the output something you can actually send.
The habit to carry: before hitting send on a prompt, read it back and ask “what is this making the model guess?” Every guess you remove is one less way the answer can miss.