Prompt engineering for Microsoft Copilot: techniques for better answers
João Barros
07 de November de 2025
1 min read
Prompt engineering is the most valuable skill for getting the most out of Microsoft Copilot. It is not about "tricks" — it is a systematic method for clearly communicating intent to the language model.
Anatomy of a good prompt
Components:
1. ROLE — who Copilot is in this context
2. TASK — what it should do (specific action verb)
3. CONTEXT — relevant data, constraints, audience
4. FORMAT — how it should present the answer
5. TONE — communication style
Weak example: "Summarize this meeting"
Strong example:
"Act as chief of staff. Summarize the attached board meeting
in 5 executive bullet points. For each point include: decision,
owner and deadline. Use direct language, no jargon. Markdown
format. Audience: CEO with no technical context."
Few-shot prompting — teaching with examples
Prompt:
"Classify support tickets as: URGENT, NORMAL, LOW.
Urgency criterion: production impact or total blockage.
Examples:
'Billing system inaccessible since this morning' → URGENT
'Update the logo on the internal portal' → LOW
'Monthly report shows yesterday's data instead of today's' → NORMAL
Now classify: 'User cannot log in since yesterday's
update — affects 30 people in the finance department'"
Chain-of-thought — step-by-step reasoning
Prompt:
"Analyze Q1 vs Q2 sales in the attached data. Think step by step:
1. Identify the total variation
2. Identify the product/region with the highest growth
3. Identify the biggest decline
4. Propose 2 concrete actions based on the analysis"
Prompts for Power BI Copilot
// More effective:
"Create a choropleth map visual showing revenue by district
of Portugal, with a blue color scale, for the year selected in the filter"
// Less effective:
"Make a sales map"
Conclusion
Investing 2 minutes in a well-structured prompt saves 20 minutes of iterations. Role + task + context + format is the base formula. For complex analyses, add chain-of-thought by asking for explicit reasoning before the final answer.