How to Write Better Prompts for Everyday Work Tasks

Most disappointing AI output traces back to a thin prompt. "Write an email to a customer about a delay" gives you something that reads like a form letter. This guide shows you a simple structure, then walks through before and after examples for common work tasks. You finish with a reusable template and a habit for improving a prompt in two or three rounds. The craft has a name, prompt engineering, and the glossary has an entry on prompt engineering, but for office work it comes down to clear writing. Which tool you use matters less than the prompt, and ChatGPT vs Claude for business covers how they differ.

The five parts of a good work prompt

Include these five things, in roughly this order:

1. Role or situation. Who is the AI helping, and in what setting? "You are helping the office manager of a small Sydney electrical firm."
2. Task. What exactly do you want produced? "Draft a reply to this customer."
3. Context. Facts the AI needs: who the audience is, what happened, what you want them to do next.
4. Constraints. Length, tone, format, things to avoid.
5. Example or format. What good looks like, or the structure you want back.

You will not need all five every time. Short, routine tasks need two or three. Anything customer-facing or important benefits from all five.

Example 1: A customer email

Before

Write an email to a customer about a delay.

The result is a generic apology full of corporate phrasing and invented details.

After

You are helping me, the owner of a small bathroom renovation business in Sydney. Write a reply to a customer whose tiling start date has moved from Monday 17th to Monday 24th because our tiler is finishing another job late. The customer is mildly annoyed and has a family event the following weekend. Tone: warm, direct, no corporate language. Apologise once, state the new date, offer a call if the new date causes a problem. Under 120 words. Do not promise compensation.

The second version names the situation, the facts, the tone, the length and a boundary. The draft needs only light editing. For the wider question, see can AI answer customer emails.

Example 2: Summarising a long document

Before

Summarise this.

After

Summarise the attached supplier contract for me, an owner who is not a lawyer. List: (1) the term and renewal date, (2) price and how it can change, (3) how to cancel and any notice period, (4) anything unusual that I should ask about. Use plain English and short bullets. If something is unclear in the document, say so instead of guessing.

The last sentence matters. Asking the tool to flag uncertainty reduces confident guessing. For legal or financial documents, treat the output as a reading aid and have a professional check anything that matters.

Example 3: A weekly report

Before

Make a report from these notes.

After

Turn the rough notes below into a one-page weekly update for our two directors. Structure: What got done, What is stuck and why, Decisions needed from you, Next week. Keep each section to three bullets or fewer. Use the numbers exactly as written in my notes and do not add any figures. Notes: [paste notes]

"Use the numbers exactly as written" guards against the tool inventing figures.

Example 4: Brainstorming

Before

Give me ideas for marketing.

After

We are a three-person bookkeeping firm in Parramatta serving tradies and small cafes. Our main work comes from referrals. Suggest 10 low-cost ways to get more referrals from existing clients, ranked by effort. For each, give one sentence on how to start this week. Avoid anything that needs a big ad budget.

Specifics about who you are and what you can afford change the quality of the ideas more than any clever wording.

Technique: ask for questions first

When a task is complex, tell the tool to ask before it writes:

Before drafting, ask me up to five questions you need answered to do this well.

You answer, and the draft improves. It also shows you what information you forgot to provide.

Technique: give it a sample of your voice

Paste two or three emails you wrote that sound like you. Add: "Match the tone and sentence length of these examples." This is more reliable than describing your style in adjectives.

Technique: iterate in short rounds

Treat the first answer as a draft. Respond with specific edits:

• "Shorter. Cut the first paragraph."
• "Less formal. Remove 'I hope this email finds you well'."
• "Make the next step clearer. One action only."

Three short corrections usually beat one long perfect prompt. When you find a prompt that works, save it. Your team can share a library, as described in our guide to rolling out AI tools to a team.

What to leave out of a prompt

• Passwords, tax file numbers and bank details.
• Customer personal information, unless your policy and the tool's terms allow it.
• Confidential client material under a contract that forbids sharing.

Check the current terms of the plan you use for how inputs are handled. Free and paid plans can differ. OpenAI's enterprise privacy page and the Anthropic docs show how those vendors describe their own handling. Your AI use policy should set the rule.

Check before you send

AI output can contain wrong facts, invented references (a failure known as hallucination, explained in what is AI hallucination) and a tone that does not match your business. Before anything goes to a customer, check names, dates, prices and promises. Our guide to keeping a human in the loop on AI output sets out a simple review routine.

Copyable prompt template

ROLE: I am [your role] at [business type, location].
TASK: [Draft / summarise / rewrite / list] [what].
CONTEXT: [Audience, situation, key facts, what I want them to do].
CONSTRAINTS: [Length, tone, things to avoid, use only the facts I give].
FORMAT: [Email / bullets / table / headings].
EXAMPLE: [Paste a sample of the style or structure, if you have one].
If anything is unclear, ask me questions before you write.

What to do next

Good prompts help one person. Getting a whole team to use them takes a plan built around the jobs your people do. AI adoption from 21Labs covers choosing the jobs, training around real work and keeping the habit going. Start with the free discovery on the form below and we reply within 24 hours.