How to ask AI for useful work
What you'll get out of thisLearner writes one role-specific business prompt that a manager judges useful.
Lesson
The single biggest mistake people make with AI is asking for the answer without giving the AI any context. The same question with two sentences of context produces a dramatically better result than the question alone.
The four-part business prompt
Every useful prompt has four parts. You don't have to write them in this order, but they all need to be there.
1. Role. Who is the AI being right now? "You are an experienced operations manager." "Act as a friendly customer service rep." "You're a careful proofreader." This sets tone, depth, and judgment.
2. Task. What specifically do you want? "Draft a follow-up email." "Summarize this thread." "Rewrite this paragraph in plain language." Be a verb away from useful.
3. Context. What does the AI need to know to do this well? Who is the audience, what's the goal, what's been said before, what tone is right? This is where most prompts fail.
4. Constraints. Length, format, things to avoid, things to include. "Keep it under 150 words. Do not use any of the words on our company's brand voice ban list. End with a clear next step."
Example: weak prompt vs strong prompt
Weak prompt:
"Write a follow-up email."
What you'll get back: a generic email that could be from anyone, to anyone, about anything. Useless.
Strong prompt:
"You are a service coordinator at a tree-removal company. Draft a follow-up email to a residential customer who received a quote one week ago and hasn't responded. The customer asked about price and timeline; we're a slightly higher quote than competitors but include cleanup. Keep it under 100 words. Friendly tone, no high-pressure language. End by offering a 10-minute phone call this week."
What you'll get back: something actually usable as a draft.
The "give it a job description" pattern
Imagine you're hiring an intern for one task. You'd write:
- ·What's the role?
- ·What's the specific task?
- ·What context do they need?
- ·What does "done" look like?
That's a prompt. The exact same shape.
What to do when the first answer isn't useful
Three moves, in order:
1. Add context. "I should have mentioned: the customer is a long-time client we don't want to lose." Most weak first answers come from missing context. Add it and re-ask.
2. Constrain harder. "Cut it in half." "Make it warmer." "Remove anything that sounds salesy." Be specific.
3. Show an example. "Closer to this tone: [paste a previous email that worked]." Examples beat instructions when instructions don't land.
Things to avoid
- ·Don't ask for facts you can't verify. "What's the population of [our city] as of last year?" The number will sound right and may be wrong.
- ·Don't ask for opinions about people. "Is this employee performing well based on [data]?" Performance evaluation is your job, not the AI's.
- ·Don't ask for one-shot legal, medical, or financial advice. "Should we sign this contract?" "Should this employee come back to work?" Wrong tool. Use AI to help you read the document. Use a credentialed human for the decision.
In-module exercise (15 minutes)
Write one prompt for a real task in your role, with all four parts:
Role:
Task:
Context:
Constraints:
Then test it (if you have AI access) or save it for the workshop. Bring it to your manager for the "is this useful?" judgment that closes Module 0.6.
Knowledge check
- ·The four parts of a strong prompt are: (a) role, task, context, constraints; (b) please, thank you, summary, format)
- ·When the first answer isn't useful, the first move is: (a) try a different AI; (b) add context and re-ask)
- ·You should NOT ask AI: (a) to draft an email; (b) to make a hiring decision)
Answers: 1. (a). 2. (b). 3. (b).
What's next
Module 0.5 teaches you to spot when AI is wrong. The three failure modes you'll see most often.