If you’ve ever typed something into Microsoft Copilot and watched it produce something wildly different from what you hoped for, you’re not alone and you’re not imagining it.
For many people working in organisations that are adopting AI technologies, you’ve been asked to use Copilot as your go-to AI platform of choice. Copilot is great but it behaves differently from ChatGPT. And if you’ve been using ChatGPT outside of work, we recommend you rethink how you prompt Copilot. You can use exactly the same prompt in both tools… and Copilot will reply like a polite intern who skim-read the brief on a moving train. Copilot is great, but it behaves differently from ChatGPT.
This is where Copilot Prompt Engineering comes into its own.
When you give Copilot a clear brief and a bit of structure, it goes from “helpful but confused” to impressive. What you need to do is show it what good looks like so it can deliver.
Here’s how you can do just that.
Why Does Copilot Need a Different Approach to ChatGPT?
Copilot is integrated inside your work. It’s sitting in Outlook, Word, Teams, PowerPoint, Excel… and it’s hoovering up whatever context is in the document or thread. That means two things:
- It makes assumptions you didn’t ask for.
- It responds more literally than ChatGPT does.
Microsoft’s own Copilot guidance makes this clear: the tool follows your instructions literally. If the prompt is vague, Copilot doesn’t fill in the blanks or make creative leaps – it plays it safe and produces the most neutral, middle-of-the-road output it can. In other words, the quality of the brief directly determines the quality of the result. So if you’ve ever thought:
- “Why is Copilot so formal?”
- “Why did it ignore half my request?”
- “Why does it sometimes sound like a GDPR policy?”
…it’s not you. It’s the prompt.
What Exactly is Copilot Prompt Engineering?
Consider Copilot prompt engineering to be more akin to briefing a colleague than making a command.
It’s a way of structuring your request so Copilot knows:
- The role it’s meant to play (analyst, comms partner, coach, project manager…)
- Who the message is for
- The outcome you want
- The format you need
- What “good” actually looks like
Including these elements in your request means that you will notice an immediate acceleration in the quality of your output.
If you are struggling for ideas of where to start with your Microsoft Copilot prompting, this blog will give you a great starting point.
Which Copilot Prompt Engineering Frameworks Actually Work?
There are loads of prompt structures out there, and most of them follow the same principles. Most importantly, be clear about the role Copilot is playing, the goal, the context, and the shape of the output. At VTT we’ve explored them all, and the good news is you don’t need to memorise a dozen acronyms to get great results. They all point in the same direction.
To keep this blog focused, we’re using ROSES as the example. It’s simple, practical and works brilliantly with Copilot because it gives your prompt enough structure to stop Copilot guessing and start producing something genuinely useful.
How does ROSES help Copilot stop guessing?
ROSES gives your prompt the structure Copilot needs:
- Role – Who is Copilot being?
- Objective – What is the single goal?
- Scenario – What’s the context or constraint?
- Expected solution – What should the output look like?
- Steps – Any process you want it to follow.
Example before / after
Vague: “Write an update on the project.”
ROSES version:
“Role: Act as my internal comms partner.
Objective: Create a concise update on Project Atlas.
Scenario: The exec team needs a straight-talking summary with no jargon.
Expected output: 5 bullet points covering progress, risks, decisions needed, next steps, and one recommended action.
Follow this order.”
Ultimately, the more context and specifics you give Copilot, the more useful and targeted your output will be.
Remember, even the best prompts sometimes create outputs that have AI hallucinations. Please remember to engage your brain with some good old fashioned critical thinking.
How does Copilot Prompt Engineering improve real work?
Here are some great examples of where you can tweak your prompt in order to uplevel your output.
Example 1: Email rewrite
Before: “Rewrite this email.”
After:
“Act as my comms partner.
Rewrite this email for a manager who’s in a hurry.
Tone: warm, human, concise.
Format: 3 short paragraphs max.
Outcome: clarity, not politeness.”
Result: a readable email.
Example 2: Exec summary
Before: “Summarise this.”
After:
“Summarise for execs.
Format:
- Progress (3 bullets)
- Risks (2 bullets, bold the titles)
- Decisions required (1 sentence)
Keep it sharp.”
Result: something you could actually use.
Example 3: Agenda creation
Before: “Make an agenda.”
After: “Create a 45-minute session agenda.
Include 6 segments, one interactive moment, one decision point, and clear timings.”
Result: a usable agenda, rather than a list of text.
What Mistakes Do People Make With Copilot (And How Do We Fix Them)?
Let’s call them out.
Mistake 1: Being vague
For example – typing, “Improve this.”
Fix: say how to improve it (shorter, clearer, warmer, punchier).
Mistake 2: Not naming the audience
If you don’t specify who your target audience is, Copilot will assume corporate formality.
Fix: use a persona or example colleague to ask it to output for.
Mistake 3: Giving no format
Copilot fills in the blanks with whatever style Outlook likes.
Fix: specify the structure or sequence you want to see.
Mistake 4: Treating Copilot like ChatGPT
Copilot and ChatGPT are different tools – they have different behaviours.
Fix: use ROSES or a similar model that is clearer on the context and specifics. I find talking to as if it’s a slightly inexperienced coworker rather than ChatGPT style informed intern really helps.
Copilot Prompt Engineering: Quick Prompt Patterns You Can Steal
Here are some Copilot-friendly prompt patterns which are short, simple, and really effective.
For exec summaries:
“Summarise for execs. Format: headline + 3 bullets (progress, risks, decisions). Tone: direct.”
For email rewrites:
“Rewrite for a time-poor manager. Tone: warm and concise. Max 120 words.”
For slide-ready content:
“Condense into a one-slide summary: headline, 3 supporting bullets, 1 recommendation.”
For meeting agendas:
“Create a 45-minute agenda with 5 segments, timings, and one decision point.”
For refining messy drafts:
“Tighten this by 40 percent. Make it conversational. Highlight anything unclear.”
Final Thought: Copilot Is Just A Colleague You Haven’t Trained Yet
Copilot Prompt Engineering helps make Copilot’s output more predictable, and more genuinely helpful. It can give you:
- Better first drafts
- Fewer rewrites
- Clearer communication
- Faster output
- Less cognitive drain
Copilot just needs clarity, structure, and a prompt that actually tells it what you want.
If your teams want to get better at this, check out our Prompt Engineering Fundamentals workshop.
Want to learn more about introducing prompt engineering to your teams?
Get in touch with our dedicated learning advisors to find out how we can help.
