Prompt engineering examples for work

The pattern behind every good work prompt

Most strong prompts contain the same five parts. You will see them in every example below:

• Role and audience: who the output is for.
• Task: exactly what to produce.
• Context: the facts, notes or data it needs. Never assume it knows.
• Constraints: length, tone, format, what to avoid.
• Check: ask it to flag assumptions or missing information.

1. Turning rough notes into a clear email

Why it works: it names the reader, the single purpose, the tone and a word limit. Asking for questions first stops it from inventing details such as the reason for the delay.

Write an email to a client who is a busy operations director. Purpose: confirm the delivery date has moved from 3 March to 10 March because a supplier is late. Tone: plain, apologetic but not grovelling, British English. Under 120 words. Include one sentence on what we are doing to prevent it happening again. Here are my notes: [notes]. If anything in my notes is unclear, list your questions before the draft.

2. Meeting notes to actions

Why it works: "only include" and "not stated" reduce invented actions, the most common failure with meeting summaries.

From the transcript below, list every action as a table with columns: action, owner, deadline. Only include actions someone actually agreed to. If an owner or deadline was not stated, write "not stated" rather than guessing. Then give a three-line summary of decisions made. Transcript: [text]

3. A first draft of a report section

Why it works: limiting it to your material keeps it accurate. The final request makes it do some thinking you can then challenge.

You are helping me draft the "Risks" section of a quarterly report for a non-technical board. Using only the bullet points below, write 200–250 words in short paragraphs. Do not add risks that are not in my list. End with the two risks you think need the most board attention and say why, based only on what I have given you. Bullets: [bullets]

4. Spreadsheet formulas you can trust

Why it works: describing the layout precisely gets a correct formula. The explanation and manual check mean you understand it and can verify it, rather than pasting it blindly.

In Excel, column A has dates, column B has sales values and column C has region names. I want total sales for the "North" region in March 2026. Give me the formula, explain each part in one line, and give me a quick way to check the answer by hand.

5. Research summaries with sources

Why it works: AI tools can state wrong facts confidently. Asking for sources and "verify" flags tells you what to check. Always check before relying on it, especially on UK rules, which change.

Summarise the main differences between [option A] and [option B] for a small UK business. Use a table. For every factual claim, say where it comes from, and mark anything you are unsure of as "verify". Do not give legal or tax advice; flag where a professional should be consulted.

6. Feedback on your own writing

Why it works: giving the AI a critical role produces more useful feedback than "make this better". Limiting it to five points keeps it actionable.

Act as a sceptical reader in procurement. Read my proposal below and list the five points most likely to make you say no, in order of seriousness. For each, suggest a specific fix. Do not rewrite the whole proposal. [proposal]

7. Customer reply templates

Why it works: templates with marked gaps keep a human in the loop, which customers notice and appreciate.

Create three reply templates for customers asking for a refund outside our 30-day window: one where we can offer a partial credit, one where we cannot help, and one that escalates to a manager. Friendly, clear, no jargon, under 100 words each. Leave [square brackets] wherever a person must personalise the reply.

Iterate instead of starting again

The first answer is a draft. Improve it with short follow-ups: "shorter", "more formal", "you assumed X, but actually Y", "give me two alternatives for the opening". This is usually faster than rewriting the whole prompt, and it teaches you which instructions matter for your work.

When a prompt works well, save it in a document with notes on when to use it. A small personal library of tested prompts is one of the most valuable AI skills you can build, and it is good evidence for employers.

Check your skills

If you want an independent view of how well you prompt, the free PromptRank assessment is marked by a fixed rubric (no AI marking), so the same answers always get the same score.