Phase 2 · Microsoft Copilot · Level 2 · Practitioner
Pages, Notebooks and agents: Copilot beyond the chat box
By the end, you'll be able to…
- Choose between Copilot Pages and Notebooks for turning chat into lasting, shared work
- Say what the Researcher and Analyst agents do and when each is the right tool
- Pick a model in Copilot Chat, and know that non-OpenAI models such as Claude can be selected
Why it matters
A chat window is great for a quick answer but hopeless as a place to build something over days or with a team. Microsoft 365 Copilot now wraps the chat in surfaces and agents that fix that: Pages you can co-edit, Notebooks that stay grounded in a curated set of sources, and specialist agents that research or analyse for you. This lesson maps them so you know which to reach for, and shows that you can even choose which underlying model does the thinking.
Why a chat box isn't enough
The core Copilot loop of ask, read, refine is perfect for a one-off. But real work is rarely one-off. You want to keep a good answer, build on it tomorrow, share it with a colleague, or point Copilot at only the five documents that matter rather than your whole digital life. That's what this lesson's features are for. Most sit on top of the same Microsoft 365 Copilot licence you've met throughout this phase, so, once more, they generally need that paid licence. There is one notable exception now, and it's good news: Copilot Notebooks have opened up to Copilot Chat users too, not just fully licensed ones, which we flag in the Notebooks section below. Here's the map.
Pages: turn a good answer into shared, editable work
Copilot Pages is an interactive canvas that turns a Copilot chat response into content you can keep and co-edit. When Copilot gives you something worth holding on to, you send it to a Page, where it becomes an editable document living alongside the chat. You can refine it with more prompts, edit it by hand, and, crucially, share it so colleagues edit it with you in real time, the way you'd collaborate in a shared document.
The mental shift: chat is ephemeral and private to you; a Page is durable and shareable. Use a Page the moment an answer stops being "a thing I asked" and becomes "a thing we're working on".
Draft a project one-pager for our new customer-feedback process, covering background, objective, deliverables, timeline and owners, as a starting point I can move to a Page and develop with the team.
Why this works: It asks for a first draft you'll then move to a Page and develop with others, framing the output as the start of shared work, not a throwaway answer. Naming the structure makes it a better seed to build on.
Notebooks: a workspace grounded in sources you choose
Copilot Notebooks solves a different problem: focus. A Notebook is a workspace where you gather a curated set of sources (Word documents, PowerPoint decks, Excel sheets, Pages, links, notes) and Copilot then answers using only that content. Instead of reasoning over everything you can access, it reasons over the handful of things you deliberately put in front of it.
That grounding on a fixed set is what makes Notebooks reliable for a project. Ask a question and the answer is drawn from your sources, not the whole web or your entire mailbox, so it's more predictable, and easier to trust, than an open-ended chat. Think of a Notebook as "Copilot, but only about this".
The access news here is worth calling out: Copilot Notebooks are no longer limited to people with a full Microsoft 365 Copilot licence. They have opened up to Copilot Chat users too, so you can now create a Notebook, add sources and ground answers on them without the paid add-on. That makes the exercise below something most people can actually try, not just read about.
Using the sources in this notebook (the project brief, the ops meeting notes and the sales data), what are the three biggest risks to the feedback project actually landing, and what does each source say about them?
Why this works: It relies on the Notebook already containing the brief, the meeting notes and the sales data, and asks a question that spans all three, exactly what Notebooks are for. Because the sources are fixed and known, the answer is easy to verify against them.
Pages or Notebooks?
A quick rule: reach for a Page when you have one thing to produce and share (a note, a plan, a draft), and a Notebook when you have several sources to reason over repeatedly (a project you'll keep asking questions about). They work together: a Page can be a source inside a Notebook.
Agents: Researcher and Analyst
Beyond surfaces, Copilot includes agents: assistants built for a specific kind of multi-step job, that do more than answer a single prompt. Two built by Microsoft are worth knowing now.
- Researcher tackles complex, multi-step research and returns a structured, source-cited report. It can gather and synthesise from the web and from your own files, emails, meetings and chats (within your permissions), working through a question the way a research assistant would rather than answering in one shot. Use it when a question needs digging across many sources, not a quick reply.
- Analyst is a data-analysis assistant, like having a data analyst on hand. It can work through data step by step (it can even run code to crunch numbers), making sense of a dataset without you needing the expertise yourself. Use it when the job is understand this data, beyond what a single Excel formula gives you.
Research how comparable mid-size UK firms handle customer-feedback tracking: common approaches, typical pitfalls, and what 'good' looks like. Give me a structured, cited summary I can share with the sponsor.
Why this works: It sets a multi-step research task, synthesising across sources with a cited output, which is what Researcher is built for and what a plain chat prompt handles poorly. Asking for citations lets you check its reasoning.
Agents are powerful, which is exactly why the over-trust rule bites hardest here: a Researcher report looks authoritative (structured, cited, confident) and can still contain a wrong or misread source. The citations are there so you can check; a cited claim is an invitation to verify, not a guarantee.
Choosing the model
Here's something many people don't realise: in several Copilot experiences you can choose which underlying model does the work. Options typically include Auto (Copilot picks), a ChatGPT model from OpenAI, and, yes, a Claude model from Anthropic. Copilot is not locked to a single vendor's models; non-OpenAI models are selectable in places like Edit with Copilot in the Office apps and the Researcher agent.
Why care? Different models have different strengths and styles, so if one's answer feels off, switching can help. One UK-specific caveat to hold: Anthropic (Claude) models are controlled by an admin setting, and for UK and EU customers they're often off by default, so if you don't see Claude offered, your organisation may simply not have turned it on. That's an administrator's choice, not a limitation of your account.
Try it now
Common mistakes
- Using chat for work that should live in a Page. If an answer is worth keeping or sharing, move it to a Page. Leaving it in chat means it's lost the next day and no one else can build on it.
- Expecting a Notebook to know things you didn't add. A Notebook answers from its sources. If you didn't put a document in, it isn't in scope; thin answers often mean a missing source, not a weak tool.
- Reaching for an agent when a prompt would do. Researcher and Analyst are for real multi-step jobs. For a quick summary or a single formula, ordinary Copilot is faster.
- Assuming Copilot only uses OpenAI models. Non-OpenAI models, including Claude, are selectable in several experiences, though in the UK Claude may be off until an admin enables it. Not seeing it isn't a fault.
- Over-trusting an agent's polished report. A Researcher report's structure and citations make it look verified, but a cited source can be misread and a confident conclusion can be wrong. The citations exist so you can check; follow the ones a decision rests on rather than trusting the summary because it reads well.
Keeping current
Pages, Notebooks and the agent line-up are among the fastest-moving parts of Copilot, and model-choice availability shifts by region and admin setting. Treat specifics as a snapshot. A few current threads to watch: multi-model selection is spreading across more Copilot surfaces, and Claude is now selectable in Copilot Chat itself (not just in the Office apps), still subject to the same admin setting that leaves it off by default for many UK and EU tenants. Microsoft is also previewing Microsoft Scout, an always-on personal agent that works in the background under its own identity; it is early-stage and preview-only, so treat it as one to keep an eye on rather than rely on. For current detail see Microsoft's Get started with Microsoft 365 Copilot Notebooks, Get started with Microsoft 365 Copilot Pages and Agents built by Microsoft. Accurate as of 14 July 2026.