Researcher in Microsoft 365 Copilot: How Multi-Model Deep Research Works

Every knowledge worker has the same problem: there’s far more information than anyone can process, and the gap between raw information and a decision is where the hours disappear. Analyst reports, internal strategy decks, email threads, competitor websites. Pulling it together into something a board can act on used to take days.

Researcher is the Microsoft 365 Copilot agent built for exactly this kind of deep, multi-step research. And in 2026 it became multi-model: it can now use OpenAI’s GPT models and Anthropic’s Claude models together. Here’s what that means, how I use it, and how to get the best out of it.

What Researcher is

Researcher is an agent in Microsoft 365 Copilot that gathers, analyses and condenses information from the web and your work content: documents in SharePoint and OneDrive, emails, meetings and chats, through Microsoft Graph and within your existing permissions. Instead of returning a list of links, it plans the research, works through multiple sources and produces a structured report with citations.

That’s the key difference from search. Search returns documents. Researcher returns an answer, with the sources you need to check it.

How multi-model Researcher works

Microsoft has added three multi-model capabilities to Researcher:

• Model choice: you can choose which model family Researcher uses, including GPT or Claude, for a given piece of work.
• Critique: one model drafts the research and another reviews it for accuracy, completeness and citation quality before it reaches you. Microsoft describes the initial pattern as GPT drafting and Claude reviewing, and reports a measurable improvement on deep-research benchmarks.
• Model Council: the same question goes to several deep-reasoning models at once. You see each full report, plus a summary of where they agree, where they differ, and what each adds. This is currently offered through Microsoft’s Frontier early-access programme.

Why this matters: different models have different strengths and blind spots. Having one model check another, or comparing several side by side, is a practical way to reduce the risk of a confident but one-sided answer, which is exactly what you want for strategy and risk decisions.

Example: a board paper in hours, not days

A strategy analyst is preparing a board paper on entering a new market. Traditionally that means days of pulling analyst reports, searching databases, reading competitor material and digging through internal documents.

With Researcher, the analyst asks:

“Assess the competitive landscape for enterprise HR software in the UK mid-market: the main vendors, pricing models, recent market moves and the risks for a new entrant. Include anything relevant from our internal strategy documents from the last 12 months, and cite every source.”

Researcher returns a structured report combining external and internal sources. The analyst’s role shifts from gathering to evaluating: checking the sources, spotting gaps and asking follow-up questions. The output can then go into Word for the board paper, or into Excel for the data.

Where Researcher adds the most value

• Strategy: competitive intelligence, market analysis and due diligence.
• Legal and compliance: regulatory change summaries and policy comparisons (always reviewed by a qualified professional).
• Finance: background research on companies, sectors and ESG topics.
• Sales: account research before major pitches, combining public information with internal account history.
• IT and architecture: technology option papers and vendor comparisons.

Tips for better research prompts

1. State the decision the research supports, for example “to decide whether to enter market X”.
2. Define the scope: region, time period, and which internal sources to include or exclude.
3. Ask for the format you need: executive summary, comparison table, risks, open questions.
4. Ask it to flag uncertainty and conflicting sources.
5. Use Model Council or critique for high-stakes questions where you want a second opinion.
6. Always check the citations before anything goes to the board.

For admins: what to set up

• Researcher is part of Microsoft 365 Copilot. See Copilot licensing explained.
• To use Claude models, an administrator must enable Anthropic models in the Microsoft 365 admin centre. Review this with your data protection and compliance teams first.
• Researcher works within existing permissions, so oversharing directly affects what it can include. Clean that up first (see overshared SharePoint), and keep content current so research isn’t built on stale documents.

FAQ

Is Researcher the same as Copilot Chat?

No. Copilot Chat is optimised for quick answers. Researcher is an agent for deeper, multi-step research that takes longer and produces a structured, cited report.

Does Researcher use my company’s data?

It can, through Microsoft Graph, and only content the user already has permission to access. You can also steer it towards or away from internal sources in your prompt.

Which AI models does Researcher use?

Researcher can use OpenAI GPT models and Anthropic Claude models, including model choice, a critique pattern where one model reviews another, and Model Council for side-by-side comparison.