Before I Ask Copilot to Write Anything, I Give It These 5 Things

The fastest way to get a generic answer from Copilot is to give it a generic request.

"Write an email about the project" technically describes a task. It does not tell the tool what happened, who needs the information, what decision you want, or what the email should avoid. Copilot fills those gaps with assumptions, and then you spend your time correcting a draft that never had enough information to succeed.

Before I ask Copilot to write anything, I give it five things: context, audience, goal, source material, and constraints. The quality difference is usually much larger than the extra two minutes this takes.

1. Context: what is happening and why does this exist?

Context gives the draft a reason to exist. I explain the situation, what led to it, and what the reader already knows. I do not paste confidential material into a tool that is not approved for it, and I follow my workplace's rules about which systems and data I can use.

For example: "We completed the first phase of the rollout. The leadership team already knows the original timeline, but they have not seen the results or the two issues that could affect phase two." That is much more useful than "This is a project update."

2. Audience: who will read it and what do they care about?

The same information should sound different in a note to a close project partner, an executive update, and a request to another team. I tell Copilot who the reader is, how familiar they are with the topic, and what matters most to them.

This does not mean stereotyping the audience. It means choosing the right level of detail. A technical partner may need the mechanism. A senior leader may need the decision, business impact, risk, and recommendation.

3. Goal: what should happen after they read it?

A draft gets much clearer when I name the desired outcome. Do I want approval, awareness, feedback, a decision, or a specific action by a particular date?

"Inform the team" often produces a soft summary. "Ask the team to choose between these two options by Thursday so we can keep the launch date" gives the message a spine.

4. Source material: what facts must the draft use?

I give Copilot the notes, facts, approved language, and key points it should work from. If the source is long, I may first ask it to organize the material into facts, decisions, risks, and open questions. Then I check that structure before asking for prose.

AI is much better at shaping useful source material than inventing the missing substance. My judgment still determines what is true, relevant, and safe to share.

5. Constraints: what should the result sound like and avoid?

Constraints prevent the most predictable cleanup. I specify the format, approximate length, tone, required details, and anything the draft should not do.

For example: "Keep this under 200 words. Put the decision request in the first paragraph. Use plain language. Do not add facts or make the risk sound resolved. End with the two options and the date we need a response."

A weak prompt versus a useful prompt

Weak: Write a professional email asking for a decision.

Stronger: Draft an email to three cross-functional partners who understand the project but have not seen this week's testing results. We need them to choose Option A or Option B by Thursday. Use only the notes below. Lead with the decision, summarize the tradeoff in plain language, keep the email under 180 words, and flag any missing fact instead of filling it in.

The stronger version is not complicated. It simply gives the tool the information a capable human collaborator would need.

The final review still belongs to you

Before I use any AI-assisted draft, I check the facts, audience, tone, missing context, and whether the message makes the next action obvious. I also remove language that sounds polished but says very little.

Copilot can help me move faster from raw material to a first draft. It cannot own the judgment, relationship, or consequences behind the message. That is still the work.

For more examples of where AI is useful in a real corporate week, visit AI at Work.