Think Together: Pages and Notebooks

Keep useful Copilot work where others can find it. Put one response on an editable Page, or use a Notebook when the work depends on a source set.


What you'll learn

  • Decide when a task calls for a Page, a Notebook, or Loop
  • Turn a Copilot response into an organized, shareable Page
  • Ground a Notebook and extract facts with evidence labels
  • Verify a generated draft against its sources before sharing
On this page

Copilot has compared three vendor quotes and given you a useful summary. Twenty minutes later, a colleague needs it. The answer is now somewhere in the chat history, awkward to find and harder to share. Pages and Notebooks give that work somewhere to live.

A Page starts with an answer. A Notebook starts with sources

A Copilot Page is an editable canvas created from a Copilot response. Move a useful answer out of chat, revise it, add your own notes, and share it as a planning brief, comparison, or status update. The route is simple: prompt, response, Page, edit, share.

A Copilot Notebook begins with a collection of material. You add files, meeting notes, past chats, and other Pages, then ask Copilot to work only from that set. Use it when the job depends on returning to the same evidence more than once.

Loop sits nearby, but it serves a different kind of work:

Surface Reach for it when the main job is...
Loop Real-time, modular collaboration on components a team co-edits all week
Copilot Pages Capturing one useful Copilot response and shaping it into a shareable artifact
Copilot Notebooks Synthesizing a bounded set of references you keep returning to

Already have the answer? Put it on a Page. If gathering and comparing sources is the difficult part, start a Notebook.

Persistence is not correctness

A Page or Notebook keeps content. Checking accuracy is a separate job. Names, dates, numbers, and recommendations still need a source check, since persistence makes a mistake easier to copy forward.

Turn one response into a Page

Start in the Microsoft 365 Copilot app with an answer worth keeping. If your prompts still come back vague, revisit the anatomy of a great prompt. Create a Page from the response, then work on it as a draft.

There are two decisions to make. First, organize the working content with useful headings, a table, or a short list of next steps. Then set the collaboration boundary by deciding who may read the Page and who needs to edit it.

Ask what each person must do. A leader who only needs the findings is a reader. A colleague correcting a number is an editor. Sharing labels vary by environment, so read what each option permits and choose the least access that still lets the person finish the work.

Put the Page's purpose at the top. Replace headings such as Numbers with headings that say what the figures mean, such as Registered and Attended. If the source doesn't explain why attendance dropped, use the exact note "The notes don't state the reason" and leave the cause open. Recalculate any percentage before using it.

Leave yourself enough to find the Page again

Before closing the Page, note a small retrieval fingerprint. Record its exact title, one distinctive source value such as "Session D: 12 registered, 4 attended", and one edit you made, such as "Owner to confirm". Match all three when you return. A familiar title by itself can lead you back to last month's version.

The planning brief you can keep· copilot-chat
Bad example

Write a summary of our Q3 workshop and tell leadership what to do next.

Good example

Create a one-page planning brief for our Q3 staff workshop. Goal: help the leadership team decide which sessions need follow-up before Friday. Source: Session A had 24 registered and 20 attended. B had 18 and 9. C had 30 and 29. D had 12 and 4. The notes don't say why B and D were low. Expectations: use the headings Overview, Attendance, Risks, and Next steps. Show the figures in a table. Don't invent reasons for low attendance. Keep it under 300 words.

Why this works: A tight, four-heading brief you can lift straight onto a Page, with the unexplained low attendance kept as an open question instead of a guess.

A Notebook gives Copilot an evidence boundary

In an ordinary chat, "What can you tell me about this topic?" is open-ended. A Notebook asks "What can you determine from these sources?" Each claim should therefore lead back to a reference in the Notebook.

Give the Notebook a one-sentence scope, such as "the approved materials and decisions for Fall Orientation." Add three or four closely related references before reaching for the whole department drive. Ask for the analysis you need, then review it. A broad, mixed source set tends to produce answers that are difficult to check.

Work in three passes. Begin by extracting facts into a table without recommendations. Next, ask for a synthesis that keeps confirmed requirements separate from proposals. Last, challenge the answer by listing every owner, date, or conclusion that the references don't support.

Keep the evidence status visible:

  • Confirmed: explicitly stated in an approved source.
  • Proposed: suggested in discussion, not yet approved.
  • Inferred: an interpretation added on top of what the source directly states.
  • Unresolved: the sources don't say.

Missing values stay missing. If a note says "send the welcome message before orientation" but names no owner or date, write "Not stated" in those cells until someone supplies the information.

Make "Not stated" the default

Tell Copilot to write "Not stated" for any missing owner, date, or approval. It feels pedantic, but it keeps gaps visible instead of letting a confident sentence paper over the one fact nobody knows.

Extract an evidence table· copilot-chat
Bad example

Pull the important decisions and owners from this notebook into a table.

Good example

Using only this notebook's references, extract confirmed decisions, proposals, named owners, deadlines, and unresolved questions into a table with columns Item, Evidence status, Owner, Date, and Source. Write "Not stated" wherever a source doesn't give a value, and don't treat a proposal as confirmed unless an approved reference supports it.

Why this works: A source-attributed table where every row names its source, and missing owners and dates read "Not stated" instead of a confident guess.

Challenge what isn't supported· copilot-chat
Bad example

Check your answer and fix anything that looks unsupported.

Good example

Review your last answer. List every owner, date, requirement, or conclusion that no notebook reference explicitly supports, and name the missing evidence for each one. Where two sources disagree, restore both sides instead of picking a winner.

Why this works: A short list of unsupported claims and open conflicts that catches invented precision before it reaches a document or a deck.

Choose the artifact for the job

Once you've reviewed the Notebook, choose an output that suits what the next person needs to do:

You need to... Ask for a...
Get oriented before detailed reading Audio overview
Explain a decision in connected prose Word document
Track one action per row Excel workbook
Present a topic to an audience PowerPoint deck
Explore themes and dependencies Mind map

As of mid-2026, Notebooks can generate Word, PowerPoint, Excel, and mind-map outputs on Windows and the web, alongside audio overviews. Use an audio overview to hear the main thread and open questions. Check every exact figure, date, and commitment in the written source.

A deck, document, or narration is still a draft. Proposals can quietly become decisions, and empty cells can acquire plausible values. For each material claim, find the source line that supports it. Remove the claim if there isn't one. OneNote, Loop, and Whiteboard use the same discipline for notes and ideas.

Turn the Notebook into an action tracker· copilot-chat
Bad example

Turn this notebook into an action tracker for the team.

Good example

From the approved synthesis in this notebook, create one row per action with columns Action, Owner, Due date, Status, Dependency, Source, and Open question. Leave unsupported values blank and explain each gap in the Open question column. Don't merge two actions into a single row.

Why this works: An action tracker you can hand off, where every row traces to a source and each blank is explained rather than filled with a plausible value.

Try it yourself

Catch the contradiction

Grounding only helps if you check the result. Use these three short fictional notes: a brief, planning notes, and a kickoff transcript. Brief: "The pilot includes Support and Operations. Security review must finish before invitations are sent." Planning notes: "Proposal: hold training on September 12. No owner has been assigned to access approval." Transcript: "September 12 is a placeholder, not an approved date."

  1. 01

    In the Microsoft 365 Copilot app, create a Notebook and add the three notes as references. (No access? Paste all three into one document and reason from that.)

  2. 02

    Ask for an extract table with columns Item, Evidence status, Owner, Date, and Source, using "Not stated" for any missing value.

  3. 03

    Check the training-date row. The transcript contradicts the planning note, so its status is Proposed.

    Hint: A date that one source proposes and another calls a placeholder is never "Confirmed."

  4. 04

    Check the access-approval row. No source names an owner, so Owner reads "Not stated" and the status is Unresolved.

  5. 05

    Run the challenge prompt and confirm Copilot flags any owner or approval no source supports.

An extract table where September 12 reads as a proposal, the access-approval owner reads "Not stated," and every invented detail gets caught.

Key takeaways

  • A Page captures one useful response as an editable, shareable canvas. A Notebook grounds Copilot in a bounded set of sources you chose.
  • Grounding turns "what can you say about this?" into "what can you determine from these sources?" You can check the second kind of answer against the sources. You can't check the first.
  • Extract, then synthesize, then challenge. The third pass is where unsupported owners and dates get caught.
  • Keep confirmed, proposed, inferred, and unresolved information visibly apart, and write "Not stated" for anything missing.
  • Match a Notebook's output to the reader's task, and treat every generated draft, even a polished one, as unverified until you check it against its source.

Check your understanding

  1. 1. You have a useful Copilot response and want to shape it into a shareable briefing. Which surface fits best?

  2. 2. What does it mean to "ground" a Copilot Notebook?

  3. 3. A planning note proposes September 12 for training. The transcript calls September 12 "a placeholder, not an approved date." How should a grounded synthesis label the training date?

  4. 4. You need to track one action per row with owner, due date, and open questions. Which Notebook output fits best?

  5. 5. An audio overview says "training is scheduled for September 12," but your notes only propose that date. What should you do?

Frequently asked questions

Terms used in this lesson

Copilot Page
A persistent, editable canvas created from a Copilot response so you can refine, organize, and share it.
Copilot Notebook
A project workspace where a bounded set of references grounds Copilot's answers and insights.
grounding
Limiting Copilot's reasoning to a source set you selected, so its general knowledge doesn't apply.
evidence status
A label (confirmed, proposed, inferred, or unresolved) that records how well a source supports a given claim.
audio overview
A spoken summary of a Notebook's content, useful for orientation before detailed reading.

Further reading