Two employees can use the same AI tool: one says "it doesn't work", the other saves an hour a day. Most of the difference isn't in the tool but in the instructions given to it, the prompt. Prompt engineering is the skill of writing instructions deliberately and systematically to get the result you want from AI.
The good news: it requires no technical knowledge. The eight techniques below can be applied at work right away.
1. Give it a role
Tell the AI who it should think like. The role shapes the tone and depth of the answer.
Weak: "Summarise this report."
Strong: "You are an experienced financial analyst. Summarise this report for presentation to the CEO."
2. Provide context
The AI doesn't know your organisation, your customer or your goal. Explain who it's for, why and under what conditions. The clearer the context, the fewer generic, shallow answers you'll get.
3. Define the task and format clearly
Say exactly what you want and what the result should look like: length, structure, tone.
"Write a 5-point summary. Keep each point to 2 sentences at most. At the end, give 3 recommendations under the heading 'Recommended actions'."
4. Show an example
One or two examples of the output you expect work better than long explanations. This noticeably improves results for corporate tone, template emails or classification tasks.
5. Work step by step
Instead of asking for complex work in one go, break it into steps: first analysis, then recommendations, then the text. Or ask the AI to explain which steps it will follow before giving the result. This helps you catch errors early.
6. Set boundaries
Also say what it should not do:
- "Don't add information that isn't in the text."
- "Say 'I don't know' where you're unsure."
- "Avoid jargon; write in language the customer will understand."
Boundaries like these especially reduce the risk of made-up information (hallucination).
7. Improve through conversation
The first answer is rarely perfect. Instead of starting a new chat, give feedback: "Make it shorter", "Expand the second point with a concrete example", "Make the tone more formal." Prompting is not a one-shot activity; it's iterative.
8. Ask for verification
Ask the AI to check its own answer or show which information comes from which source. But remember: the final check is always your responsibility. Always verify figures, names and legal references.
An example that puts it all together
Role: You are an HR specialist.
Context: We're hiring a senior data analyst for a mid-sized technology company. The team works remotely.
Task: Write a job ad.
Format: A short intro paragraph; 5 bullet points each under "Responsibilities" and "What we're looking for"; friendly but professional tone.
Boundaries: Don't use discriminatory language about age, gender or marital status. Don't include salary information.
Prompt libraries: from individual skill to team productivity
The most productive teams don't keep their best prompts to themselves; they share them. A department-level prompt library helps newcomers get value quickly and standardises quality.
A word of caution
However good your prompt, you should never enter personal or confidential information into unapproved AI tools. Productivity and safety should be learned together.
In our Generative AI at Work programme, participants apply these techniques to their own workflows, and each team leaves with its own prompt library. The programme can be tailored for HR, finance, marketing, sales, legal and other departments.
Sources