Copilot Prompts for Work: Documents, Research and Meetings

Good Copilot prompts specify the work you need, the information to use and how you will judge the answer. The examples below cover Microsoft Copilot for documents, research and meetings, with checks that help you separate a useful draft from an unsupported conclusion.

Start by replacing the bracketed fields and supplying the named sources through the Copilot experience you use. A reference to a document is useful only if Copilot can actually access it. Tools like fenbs keep reusable prompts outside a single conversation, so you can preserve the version that passed review and adapt it for another supported assistant.

What should you include before asking Copilot to work?

Use four short labels: goal, context, sources and expected output. Microsoft recommends these elements in its Copilot prompting guidance↗ (opens in a new tab). You do not need an elaborate persona or a page of motivational instructions; you need enough information to make the assignment clear.

For example, a document summary for a project manager should emphasize deadlines and unresolved choices. A summary of the same document for a new employee should explain terms and background. Specify the reader rather than asking for a generally professional answer, which leaves the assistant to guess what matters.

• Goal: describe one finished result, such as a one-page decision memo.
• Context: identify the reader, the purpose and any boundary on the task.
• Sources: name the files, excerpts or public pages the answer may use.
• Review: require evidence for factual claims and a separate list of missing information.

These are workplace templates, not GitHub coding instructions. Use the separate Python examples or C++ examples for software work; our Copilot versus ChatGPT guide explains the broader product choice.

How can you draft a document without inventing facts?

Give Copilot the evidence before asking for polished prose. Say a small design studio needs a project update based on an approved scope and a progress note. Ask for a draft that distinguishes agreed work, completed work and questions for the client. That distinction matters more than choosing between a formal and friendly tone.

Goal: Draft a project update for [reader].
Context: The reader needs to understand progress and choose [decision].
Sources: Use only [approved scope] and [progress note], supplied here.
Output: A short summary, completed work, remaining work, and questions.
Keep proposals separate from approved commitments. Do not invent dates, results, owners, or quotations. Mark missing information as "Needs confirmation."
Finish with a claim-check list mapping factual statements to source sections.

Before sending, compare commitments against the approved scope. Check that a planned activity has not become a completed activity and that a tentative delivery date has not become a promise. Read quoted language in the original document, because quotation marks make a particularly strong claim about accuracy.

If the draft is too long, ask for compression after checking those facts. A shorter first attempt can hide the reasoning you need to inspect. Keep the reviewed content and ask Copilot to shorten it without changing the status, dates or meaning of any commitment.

Save the prompt with its required inputs in fenbs Skills and prompts, so the next person sees that an approved scope is part of the method rather than an optional attachment.

How can you compare research without confusing evidence and opinion?

Use a comparison question with explicit criteria. For example, you might compare three public accessibility guides for planning a website review. Give Copilot the sources and ask what each one directly supports, where they disagree and which questions remain unanswered. Avoid asking for a winner before deciding what a good answer requires.

Compare [sources] to answer [specific question] for [reader].
Use these criteria: [criterion A], [criterion B], [criterion C].
For each conclusion, identify its supporting source and section.
Separate direct source statements from your interpretation. Mark an unanswered question "Not established by these sources."
Return a comparison table, disagreements, and the next source we should inspect. Do not infer that an omitted feature is unavailable.

Open the cited material yourself. A page can mention a subject without supporting the conclusion beside its link. Check publication and revision information when it matters, and make sure the source describes the same product or policy you are evaluating. A comparison based on different editions can appear precise while answering the wrong question.

Research also needs a stopping rule. Decide whether you need enough evidence for an internal discussion or a conclusion ready for publication. Copilot can help identify the missing evidence, but a longer answer does not remove that gap. Keep uncertain points visible until someone verifies them.

Add the comparison criteria before inspecting the result. If you choose them afterward, a well-written answer can steer the decision toward whatever the assistant happened to emphasize. For the accessibility example, a useful criterion might be whether each guide explains how a person verifies keyboard navigation. Record what would count as sufficient evidence before accepting the comparison.

How can meeting prompts produce actions people actually agreed to?

Ask for decisions, proposals and actions as separate groups. A lively discussion contains suggestions that no one accepted. An accurate summary should preserve that difference and leave an owner or deadline blank when the meeting did not establish one.

Copilot in Teams can summarize discussion and suggest actions. Microsoft explains that using meeting content afterward depends on a transcript, and that available behavior also depends on meeting settings in its Teams meeting guide↗ (opens in a new tab). Check the available source before assuming an earlier discussion can be recovered.

Using [available meeting transcript or supplied notes], prepare:
1. Decisions explicitly agreed to, with supporting passages.
2. Proposals still open.
3. Actions, with owner and deadline only when stated.
4. Conflicts or unclear statements that need confirmation.
Do not turn a suggestion into an agreement. Do not assign work to someone merely because they discussed it. End with questions for the meeting organizer.

Have the organizer or responsible colleague confirm the result before treating it as the project record. If notes omit the final agreement, ask the people involved rather than asking the model to resolve the ambiguity. Keep the transcript in its approved system and save only the confirmed work information elsewhere.

What follow-up prompt makes a draft easier to review?

Ask for a focused audit instead of a vague improvement. Choose the failure you want to catch: unsupported claims, changed commitments, an unclear audience or missing questions. Reviewing all of those at once often produces a general compliment followed by minor wording changes.

Review the draft against [source material]. Do not rewrite it yet.
List factual claims that lack support, changed meanings, missing qualifications, and commitments that were not approved.
For each issue, show the affected sentence and explain the source mismatch.
Then propose the smallest correction. Separate verified problems from preferences about style.

Use the audit as a checklist, then inspect the draft yourself. The same assistant can miss its original mistake. Our prompt engineering guide covers clearer task instructions; the rest of the AI prompts collection offers templates for other kinds of work.

When you save a corrected template, include one short example of an acceptable result. A sample showing an unanswered deadline as “Needs confirmation” communicates the review standard more clearly than another instruction to be accurate.

How to do this with fenbs

1. Create a free account, then add the project whose recurring documents or research you want to improve.
2. Save one reviewed template in Skills and prompts. Include its purpose, placeholders, source requirements and the checks a person must perform.
3. Record any continuing instructions people have agreed in Decisions and rules, such as requiring confirmation before a proposal becomes a commitment.
4. Connect a supported assistant using its integration guide and ask it to read the relevant prompt and project context before working.
5. After review, update the prompt when needed and save task progress, including unresolved evidence and the next step.

fenbs is free for one person, with no credit card. It does not connect to Microsoft Teams or import your meeting transcripts. Use Microsoft Copilot inside the approved Microsoft experience for that material, and use fenbs when you need reusable instructions and a work record available to supported connected assistants.

A folder of occasional prompts may be enough if you work alone in one app. Start a shared library when colleagues repeatedly ask which version to use, which source is required or who approved the wording. Save that answer beside the prompt so reuse begins with the reviewed method.