2.3.1. Researcher and Analyst: The Prebuilt Reasoning Agents
💡 First Principle: Not every agent is one somebody built. Researcher and Analyst are Microsoft-built agents that come with a Microsoft Copilot (formerly Microsoft 365 Copilot) license and are pinned in your navigation already. You don't configure their knowledge sources, because they reach whatever you can already reach. They are the answer to "I need more than a chat reply" without anyone creating anything.
What each one is for:
| Researcher | Analyst | |
|---|---|---|
| Built for | Complex, multi-step research | Data analysis |
| Draws on | The web and your work content: files, emails, meetings and chats you can access (plus connector content) | Spreadsheets and other data sources (Excel, CSV and others), across multiple files |
| Produces | A structured, source-cited report with organized sections, charts and visuals | Statistics, trends and outliers, in a report with charts and tables |
| Typical use | "Compare our last three quarterly reports against published competitor results and tell me where we are losing share" | "Find the trends and outliers in these three sales exports and summarize them with charts" |
Researcher trades speed for depth, deliberately. A simple query may return in under five minutes, and a genuinely complex one can take 10 to 45 minutes. That is the design, not a fault: it spends the time retrieving and reasoning so the output is something you can act on and share. It may also ask clarifying questions before it starts, and answering them is how you steer the scope. Inside Researcher you can also choose a model and mode, including GPT or Claude options, a built-in Critique review step and Model Council, which runs the same question through several reasoning agents at once. Claude options appear only if your admin has allowed Anthropic models.
When to choose which:
| You need… | Use |
|---|---|
| A quick summary, a short reply, light brainstorming | Standard Copilot chat |
| A back-and-forth task without heavy analysis | Standard Copilot chat |
| Deep reasoning across multiple sources, web plus your work, in a report you can share | Researcher |
| Citations and a clean structure you can present or decide from | Researcher |
| Statistics, trends or outliers across spreadsheets and data files | Analyst |
Limits worth knowing:
- Researcher can't process images inside input documents, and Microsoft points spreadsheet-heavy tasks to Analyst instead.
- Researcher can't be customized or extended (for example in Copilot Studio): it's a first-party agent with fixed logic. Treat Analyst and Microsoft's other prebuilt agents the same way: you use them as built.
- Users can't remove or unpin Researcher, but admins can block Researcher and Analyst.
- If an admin turns off web search, Researcher uses no web data.
- Both are available in the Copilot mobile app.
Both stay inside the same boundary as the rest of Copilot. They inherit the permissions, policies and compliance commitments that already apply to your tenant. Researcher can't surface a document you couldn't open yourself: the depth changes, the access rules don't.
⚠️ Exam Trap: A scenario describing a specialized, recurring need doesn't automatically mean "build a custom agent." If the need is deep research or data analysis rather than your organization's specific knowledge and rules, the right answer is a prebuilt agent that already exists. Custom agents are for scoped, organization-specific, repeatable needs, as 6.2 describes. And Researcher taking many minutes, or asking clarifying questions first, is the feature working as intended. A distractor offering "switch to standard chat for a faster answer" trades away exactly the depth the scenario asked for.
Reflection Question: A strategy lead needs a shareable, cited competitive briefing that draws on internal sales files and public market data, and a finance colleague needs the trends and outliers across three messy spreadsheet exports. Which experience should each use, and what would be lost by sending both requests to standard Copilot chat?
