Quick answer
Copilot Notebooks collect the chats, generated outputs and reference material for a single project in one place, so Copilot can draw on that accumulated context in later work. The redesigned experience rolled out through the Microsoft 365 Copilot app during 2026. You can also ground an agent on a Notebook, which keeps its answers aligned to that project's material rather than everything you have access to.
Memory personalises responses using your chat history, so Copilot gradually learns how you work and what you tend to need. General availability began in early 2026, with a redesigned settings page that shows what Copilot has remembered and lets you control it.
Used together, they turn Copilot from a tool with amnesia into something that knows what you are working on. The catch is that both reward deliberate setup and quietly degrade if you let them accumulate by accident.
Key takeaways
- Most Copilot time is lost rebuilding context, not writing prompts.
- Notebooks hold a project's chats, outputs and references in one workspace.
- Grounding an agent on a Notebook keeps its answers inside that project.
- Memory personalises replies from your history and is controllable in settings.
- One Notebook per real project. Notebooks for vague themes become useless.
- Review what Memory has stored, because wrong assumptions persist silently.
- The benefit compounds on long running work and is negligible on one off tasks.
Watch someone use Copilot for a week and you will notice how much of their effort goes into repetition. Every new chat starts with the same three paragraphs of background: who the client is, what the project involves, what tone the organisation writes in. The tool then produces something decent, the window closes, and tomorrow the whole preamble happens again. Notebooks and Memory exist to end that, and most people have never deliberately used either.
What a Notebook is actually for
The problem it solves
A significant project generates dozens of Copilot conversations across weeks. Each one starts cold. You paste the same brief, explain the same constraints again, remind it of the same audience. That preamble is often longer than the actual question.
A Notebook holds that context once. It brings together the chats you have had, the outputs Copilot produced, and the reference material the work depends on. When you return, the background is already there.
What to put in one
Be selective. A Notebook works because its contents are relevant, and adding everything vaguely related defeats the purpose.
Good contents for a client project Notebook:
- The brief or scope document
- Any style or tone guidance that applies
- Key decisions made so far, and the reasoning
- The two or three source documents the work keeps returning to
- Outputs you were happy with, as examples of the standard
Avoid dumping in the entire shared drive. Precision in what a Notebook contains directly determines the quality of what comes out of it.
One Notebook per real thing
The mistake people make is creating Notebooks for broad themes: Marketing, Admin, Ideas. Those become junk drawers within a fortnight and produce vaguer answers than working without one.
Create a Notebook when there is a genuine bounded piece of work with a beginning and an end. A specific client engagement. A tender response. A funding application. A programme running over a term. When the work concludes, the Notebook stops being live and becomes a record of what was done.
Grounding an agent on one
You can point an agent at a Notebook so it draws on those references and working materials specifically. That is genuinely useful when you want consistency: an agent grounded on your tone guidance and three approved examples will produce output that matches your house style far more reliably than one working from general instructions.
This is the closest thing in Copilot to having an assistant who already knows the account.
What Memory changes
How it works in practice
Memory lets Copilot personalise responses using your chat history, so it picks up patterns in what you ask for and how you like it presented. Over time, replies arrive closer to the shape you would have asked for. Voice conversations in the Copilot app can draw on stored memories too, which makes spoken use considerably more practical.
The effect is subtle rather than dramatic. You notice it mostly as an absence: less correcting, less restating your preferences, fewer replies pitched at the wrong level.
Check what it has stored
This is the part people skip. A redesigned Memory settings page gives visibility and control over what Copilot remembers, and it is worth ten minutes of your attention.
Two things go wrong without review. First, a temporary situation becomes a permanent assumption. You spent three weeks on one client, and now Copilot treats that context as your default. Second, a preference you expressed once in a specific situation gets applied everywhere.
Neither announces itself. You simply get slightly wrong answers with no obvious cause. Reviewing stored memories periodically fixes both in minutes.
When to turn it down
If your work spans genuinely different domains, or you switch between contexts where the right tone differs sharply, heavy personalisation can hurt. The controls exist for this. It is not an all or nothing setting, and there is no virtue in leaving everything on if the results get worse.
A setup that pays for itself
- Create Notebooks only for live, bounded projects. Three or four at a time is plenty for most people.
- Put the standard in the Notebook. Examples of good output are worth more than descriptions of what good means.
- Keep decisions in there. Copilot cannot infer why you rejected the earlier approach unless you wrote it down.
- Ground your recurring agent on the relevant Notebook. Consistency improves immediately.
- Review Memory monthly. Delete anything that reflects a situation that has passed.
- Close Notebooks when work ends. They are project workspaces, not permanent archives.
That is perhaps twenty minutes of setup and ten minutes a month of maintenance, against a preamble you were otherwise retyping several times a day.
Where the benefit actually lands
| Type of work | Benefit from Notebooks and Memory |
|---|---|
| Long running client project | High. Context accumulates and stops being retyped |
| Recurring monthly reporting | High. The standard stays consistent between runs |
| Team writing with a house style | High, particularly with an agent grounded on examples |
| One off summary or quick draft | Minimal. Ordinary chat is faster |
| Exploratory research | Mixed. Useful for a theme, wasteful for a passing question |
The pattern is simple. These features reward continuity. If the work has a memory, giving Copilot one helps. If the task is finished in five minutes and never returns, the setup costs more than it saves.
Frequently asked questions
Are Notebooks the same as Copilot Pages?
They serve different purposes. Pages are collaborative documents that a conversation can produce and a team can work on together. A Notebook is a workspace that gathers the chats, outputs and references for a project so that context persists across many conversations.
Can my colleagues see my Notebooks?
Notebooks follow the normal Microsoft 365 permission model, so what others can see depends on how the contents are shared. Treat a Notebook's sensitivity as equal to the most sensitive item you put in it, and share deliberately rather than by default.
Does Memory mean Copilot is storing everything I type?
Memory personalises responses using chat history, and the settings page is designed to show you what has been retained and give you control over it. If you are unsure, review the page. Vague worry about what is stored suppresses use far more than the actual behaviour warrants.
Will this make Copilot better at my job specifically?
It makes Copilot better at the work you have given it context for. That is not the same as understanding your role. The gain comes from removing repeated explanation, not from the tool acquiring expertise.
Is it worth setting up for a small team?
Yes, and arguably more so. Small teams carry a lot of shared context informally, and grounding an agent on an agreed Notebook is a cheap way to make the house standard consistent when several people are producing work.
Author note
Written for AI Vision Consulting, a practical AI training and automation company based in Newcastle upon Tyne and serving UK organisations. Most of the value we find in Copilot comes from removing repeated effort rather than from any single impressive feature.
Sources
Related articles
More practical guides from AI Vision Consulting.
