Template · Asking well
AI prompt template for work tasks
Most weak AI output starts with a short request for a large task. The model fills each gap with a guess, and the guesses add up to a generic draft. This template puts the gaps on the page before you send anything.
Fill in each field in a sentence or two. Then paste the filled form into the conversation as your request. The same form works for drafts, summaries, comparisons, and replies.
When to use it
- A task will take the model more than one paragraph of output.
- The output goes to someone other than you.
- You have asked for this kind of work before and needed several rounds to get it right.
The template
Task
What do you want produced, in one sentence?
Reader
Who will read it, and what do they already know?
Purpose
What should the reader do or decide after reading it?
Source material
What should the model work from? Paste it below the form.
What the material means
What would a new colleague need explained about the material?
Format
What shape, length, and sections should it have?
Limits
What must it respect, and what must it leave out?
Gaps
What should the model do if something is missing?
A filled example
A project coordinator needs a short status update for department heads about a delayed office move. The facts are in a vendor email and the coordinator’s own call notes.
- Task
- Write a status update on the office move for the department heads.
- Reader
- Six department heads. They know the move was planned for the first week of March. They do not know about the delay.
- Purpose
- Each head should confirm by Friday whether their team can pack a week later than planned.
- Source material
- The furniture vendor’s email of February 12, pasted below, and my notes from the call with the building manager.
- What the material means
- “Phase 2” in the vendor email means the desks for floors three and four. The building manager’s date is fixed and cannot move.
- Format
- An email of no more than 150 words. One paragraph on what changed, one on the new dates, then the question for each head.
- Limits
- Use only the dates in the source. Do not name the vendor’s staff. Do not mention the cost dispute.
- Gaps
- If a date is not in the source, write [DATE NEEDED] instead of choosing one.
What each field is for
Task
Skill: Asking well
A one-sentence task statement keeps the request focused. If you cannot write it in one sentence, the task may need splitting into steps.
Reader
Skill: Asking well
The reader decides the level of detail and the terms to use. A model that knows the reader stops explaining what the reader already knows.
Purpose
Skill: Asking well
Purpose tells the model what the output must achieve. A draft written to get a decision reads differently from one written to inform.
Source material
Skill: Giving good context
The model has never seen your project. Pasting the source turns a plausible guess into an answer drawn from the facts.
What the material means
Skill: Giving good context
Material without meaning is half the context. One line of explanation prevents the model from misreading your own shorthand.
Format
Skill: Asking well
Format instructions are the ones a model follows most reliably. Stating them saves a round of reshaping.
Limits
Skill: Giving good context
Limits keep the output inside what you can send. Exclusions are part of good context, and the model cannot guess them.
Gaps
Skill: Asking well
Without this instruction a model fills gaps with confident inventions. A marked gap is quick to fill. An invented date can reach the reader unnoticed.
Common mistakes
Filling in the task and nothing else
The task line alone is the short request the template exists to replace. The fields below it carry the decisions the model cannot make for you.
Describing the source instead of pasting it
“See the vendor email” means nothing to a model. Paste the text, or the relevant part of it, below the form.
Using adjectives for tone
“Professional and friendly” means different things to different readers. If tone matters, add one sentence written the way you want the output to sound.