Prompt engineering: asking well

What it is

A prompt is what you type to an AI assistant, such as Claude, ChatGPT, Microsoft Copilot or Gemini. Prompt engineering simply means writing that request well, so the first answer is closer to what you need.

Think of briefing a new temp on their first morning. "Sort out the emails" gets you something. "You're covering the front desk. Reply to these three customer emails, politely, in under 100 words each, and flag anything about refunds to me" gets you something you can use. The assistant is the same: it cannot read your mind, and it fills gaps with guesses.

A good prompt can contain six parts: a role, a task, an audience, a format, examples and constraints (limits). You do not need all six every time, but the more that matters, the more you include.

Why it matters

• Clear requests save rounds of back-and-forth and rewriting.
• The assistant guesses whatever you leave out, and its guesses are often generic.
• A well-written prompt can be saved and reused by the whole team.
• Being specific about limits (length, tone, what not to include) reduces errors.
• Good prompting does not replace checking: you still review the result.

How to do it

1. Set a role: "You are an experienced [job] who writes for [type of reader]."
2. State the task in one clear sentence, with a verb: summarise, draft, compare, list.
3. Name the audience: who will read it and what they already know.
4. Specify the format: bullets or paragraphs, length, headings, a table.
5. Show an example of the style or structure you like, if you have one.
6. Add constraints: tone, words to avoid, what to leave out, "say if you are unsure".
7. Ask it to ask you questions first when the job is complex: "Before you start, ask me up to five questions."
8. Iterate: read the answer, then tell it what to change. Treat it as a conversation, not one shot.

Example

Before:

Write an email about the new expenses process.

Result: a long, generic email that could come from any company.

After:

You are an operations manager writing to a team of 30 field staff
who are busy and read email on their phones.

Task: draft a short email announcing the new expenses process,
which starts on [date].

Key points:
- Receipts must be photographed and uploaded within [number] days
- Claims over [amount] need a manager's approval
- Questions go to [contact]

Format: subject line plus no more than 120 words, three bullets,
friendly but clear. No jargon. End with one sentence on why the
change helps them.

If anything above is unclear, ask me before you write.

The second prompt takes one extra minute and usually gives a draft that needs light edits, not a rewrite. You still check dates, amounts and names before sending.

Common mistakes

• One-line prompts for important work. Fix: add audience, format and limits.
• Accepting the first answer. Fix: ask for changes: "shorter", "warmer", "remove the second paragraph", "give me three options".
• Asking for several unrelated jobs at once. Fix: one task per prompt, or number the tasks clearly.
• Hiding the real goal. Fix: say what the output is for ("this goes to the board", "this is a first rough idea").
• Using real customer details to test a prompt. Fix: test with made-up names and numbers, and use only approved tools.

Try it this week

1. Pick one email or short document you write often.
2. Write a one-line prompt for it and save the answer.
3. Rewrite the prompt using the six parts: role, task, audience, format, example, constraints.
4. Compare the two answers side by side.
5. Improve the better one with two follow-up instructions, then save the final prompt somewhere you can find it again.

Going further

• Ask the assistant to improve your prompt: "Rewrite this prompt so it is clearer, then use it."
• Use clear separators, such as headings or lines of dashes, to mark where instructions end and pasted material begins.
• Ask for several options at once ("give me three subject lines") and pick the best.
• Ask it to check its own draft against your constraints before showing it to you.
• Save prompts that work as templates, with [square brackets] for the parts that change.

Key words

• Prompt: the instruction or question you give an AI assistant.
• Role: the expert or point of view you ask the assistant to take.
• Constraint: a limit, such as length, tone or what to leave out.
• Iterate: improve something step by step through several rounds.
• Template: a saved prompt with blanks to fill in each time.

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (246 KB), or the complete AI Field Guide (1.7 MB).