How to Write Better Prompts for Work

A prompt is simply the instruction or question you give to an AI tool. It's the starting point for everything the AI produces. The quality of what you get back is directly shaped by the quality of what you put in — a vague, one-line prompt typically produces a generic, vague response, while a clear and specific prompt produces something genuinely useful.

Think of prompting like briefing a very capable but brand-new colleague. They have broad knowledge and good skills, but they don't know your organisation, your audience, your tone of voice, or the specific context of your request. The more clearly you explain these things, the more useful their output will be. This isn't complicated — it's just communication.

Strong prompts for work tasks generally include four elements: role, task, context, and format. You don't need all four every time, but thinking through each one will quickly improve your results.

• Role: Tell the AI who it should act as. "You are an experienced HR manager writing for a UK public sector audience" gives very different results from a blank prompt.
• Task: Be specific about what you want. "Write a short email" is weaker than "Write a 150-word email declining a meeting request politely and suggesting an alternative."
• Context: Provide relevant background. What organisation are you in? Who is the audience? What tone is appropriate? What has already happened?
• Format: Specify how you want the output structured. "Give me three bullet points," "Write this as a formal letter," or "Produce a table with two columns" all shape the response significantly.

Combining these elements transforms a weak prompt into a strong brief. It takes an extra 30 seconds but regularly saves you multiple rounds of editing.

Here are examples of weak prompts transformed into stronger ones for typical UK workplace scenarios:

• Weak: "Write an email about the new policy." Strong: "Write a brief, friendly internal email (under 200 words) to all staff at a local council announcing a new hybrid working policy that takes effect on 1 May. The tone should be informative but warm, and it should include a link to the full policy document and a contact name for questions."
• Weak: "Summarise this document." Strong: "Summarise the key points of this report in five bullet points for a senior manager who has 2 minutes to read. Focus on recommendations and financial implications. Avoid jargon."
• Weak: "Write a job advert." Strong: "Write a job advertisement for a Procurement Officer role at a NHS trust. The salary is £32,000–£38,000. The post is hybrid. Use a clear, inclusive tone that avoids overly corporate language. Include three bullet points of essential requirements and three for desirable ones. The advert should be under 400 words."

Notice how each strong prompt gives the AI a clear role, task, audience, and format. The AI doesn't have to guess — and you get something you can actually use.

Rarely will the first response be exactly what you need — and that's fine. Prompting is an iterative process. If the output is too long, say "Now make this 30% shorter." If the tone is wrong, say "Rewrite this in a more formal tone suitable for a chief executive." If it's missed the point, say "You've focused on X but I actually need Y — can you rewrite with that emphasis?"

You don't need to start from scratch each time. Build on previous responses in the same conversation — the AI retains the context of your earlier exchanges. You can also ask it to give you options: "Give me three different opening paragraphs and I'll choose the best one." This is particularly useful when you're unsure what tone or approach will work best.

One useful habit: ask the AI to explain its reasoning or flag any assumptions it has made. "What assumptions did you make to write this?" often reveals gaps you can then fill with better context. This back-and-forth approach is how experienced AI users work — not trying to get it perfect first time, but refining efficiently.

Not every work task is suitable for AI tools, and prompting well includes knowing when to add caution. For tasks involving personal data, legally sensitive content, or communications that could affect real individuals, be especially careful about what context you include in your prompt — and remember that some AI tools may log your inputs.

Where you're using AI to draft something sensitive — a disciplinary letter, a safeguarding communication, a response to a complaint — treat the AI output as a draft only. Have it reviewed by a qualified colleague before sending. The AI doesn't know the full human situation, and getting it wrong in these contexts can have serious consequences.

Also be mindful of confirmation bias: AI tools tend to produce confident-sounding output, which can make poor advice feel authoritative. If you're using AI to help think through a complex situation, explicitly ask it to present counterarguments or limitations: "What are the risks or downsides of this approach?" This helps you think more critically rather than simply accepting the first output you receive.