How to Write Prompts for AI Assistants That Get Useful Answers at Work

AI assistants can draft emails, summarise long documents, brainstorm ideas and explain unfamiliar topics. But many people try one vague request, get a generic answer and decide the tool isn't useful. The difference is usually in how the request is written.

Here's a simple approach that works with most AI chat tools.

Give it the context a colleague would need

Imagine asking a new colleague for help. You wouldn't just say "write an email." You'd explain who it's for, what happened and what you want to achieve. Do the same with an AI assistant:

• Who you are and what your business or role is.
• Who the audience is — a customer, your manager, a beginner.
• What's already happened — the background they can't know.

Weak: "Write an email about the delay."

Better: "I run a small printing business. A regular customer's order of 500 brochures will arrive two days late because our paper supplier was delayed. Write a short, apologetic email that explains the delay, confirms the new delivery date of Thursday, and offers free delivery as a goodwill gesture."

Be specific about the task

Say exactly what you want: summarise, rewrite, compare, list options, find problems, translate, simplify. If there are constraints, state them — word count, tone, reading level, things to avoid.

Ask for the format you need

• "Give me a table comparing the three options by price, setup time and limitations."
• "Answer in five bullet points."
• "Write it as a WhatsApp message, not an email."

Specifying the format saves you from reformatting the answer yourself.

Show an example

If you have a style you like — a previous email, a product description, a report — paste it in and say "match this tone and structure." Examples communicate style better than adjectives such as "professional" or "friendly."

Work in steps, not one giant request

For bigger tasks, break the work down:

1. Ask for an outline first.
2. Correct the outline.
3. Ask for one section at a time.

You stay in control, and each step is easier to check.

Iterate instead of starting over

If the first answer isn't right, tell the assistant what to change: "Make it shorter," "less formal," "you missed the point about pricing." Follow-up instructions usually work better than rewriting the whole request.

Always check the output

AI assistants can state incorrect facts with complete confidence, invent sources or misunderstand numbers. Treat their output as a first draft:

• Verify facts, figures, names and dates against reliable sources.
• Don't paste confidential information — client data, passwords, private documents — unless your organisation has approved the tool for that use.
• Read before you send. You're responsible for what goes out under your name.

A reusable template

Context: [who you are, the situation]Task: [what you want done]Audience: [who will read it]Format: [length, structure, tone]Example: [optional sample to match]

The bottom line

Clear context, a specific task and a defined format turn an AI assistant from a novelty into a useful tool. Check everything it produces, and let it handle the first draft while you handle the judgment.