How to write a good AI prompt for everyday work
A complete example that turns a vague request into a useful instruction for an everyday work task.
A prompt is the instruction or question you give an AI tool. To make it useful at work, explain the output you need, who it is for and which information it should rely on. A good prompt leaves less room for guesswork, but it does not remove the need to check the answer.
“Write me a professional proposal” leaves too many decisions open. AI does not know the client, the objective, the constraints or what you can actually offer. The text may look finished and still be unsuitable for the task.
Define the output before the tone
Start with the task: a follow-up email, a summary for a manager or a comparison of documents. Then add the audience and the decision the text should support. “Professional” is a useful description, but “clear to a client without technical knowledge” gives more specific direction.
Microsoft's prompting guidance distinguishes the goal, context, expectations and source. You can use this structure to check whether the tool has the information it needs without turning every request into a long form.
Add the source and the limits
When working with facts, specify which material may be used. If a deadline is missing from the notes, ask for it to be flagged for confirmation. If no price has been approved, explicitly instruct the tool not to invent one.
Another constraint might be length or format. An email may need a few paragraphs; a work plan may need a table with tasks, owners and deadlines. The format should serve the person who will use the output.
An example you can adapt
“Draft a follow-up email after a meeting with a potential client. Use only the notes I provide below. The reader is a business manager who wants to understand the next steps. Include the need we identified, what we need to clarify and a proposal for a follow-up meeting. Keep it under 180 words, in natural English. Do not invent prices, dates or commitments. At the end, list missing information separately.”
Add the notes that are approved for use after this instruction. Read the list of missing information first: it may reveal that you do not yet have enough information to send the email. The result is a draft to review, not a message to send automatically.
Improve the part that is not working
If the response is too generic, ask for a specific section to be changed. For example: “The second paragraph does not reflect the notes. Rewrite it using only the need for weekly reporting.” This gives the revision more direction than “make it better”.
If the tool repeats the mistake, review your original instruction. You may have asked for an answer that is both very short and highly detailed. Remove contradictions and specify which requirement takes priority.
Save the templates you actually use
A prompt library can start with five recurring tasks. For each template, record when to use it, which material it needs and what to check. Save the improved version after testing it in real work, rather than every variation generated during experimentation.
Frequently asked questions
Do I have to write prompts in English?
You can also work in Albanian. Assess the quality for your task and terminology; translating a prompt into English does not automatically guarantee a better result.
Is it enough to give AI an expert role?
A role can guide the perspective, but it does not replace sources, an objective or fact-checking. Supply the information a colleague would need to do the task.
Practise with a real task
In Lumina's training sessions, prompts are built around participants' work. Contact us about a programme where your team creates and tests its own templates.