M365 Copilot — Researcher Agent

At a glance

What it is
A built-in research agent that queries the web
and your M365 work data in a single request.

Key difference from Chat
Chat responds to queries against live M365 data.
Researcher is optimized for multi-step research tasks and returns cited, structured reports.

Frontier modes
Critique (Roadmap 553213): two-model review pipeline.
Council (Roadmap 558538): two models run in parallel, outputs compared.
Both require Anthropic sub-processor enabled by admin.

Use case	What Researcher does	For
Competitive landscape	Web + work data synthesis with source citations	Strategy · BD · Executive
Market or technology briefing	Structured report from web and internal context	Analyst · PM · Consultant
Pre-meeting intelligence	Who you are meeting, their org, recent news	Sales · BD · Executive
Policy or regulatory review	Find, read, and summarise relevant documents	Legal · Compliance · HSE
Internal knowledge synthesis	Pull threads from emails, meetings, files, chats	PM · Programme Director · EA

Researcher vs Chat vs Analyst — when to use each

Agent	Best for	Data sources	Output
Researcher	Multi-step research across web and organisational data — returns cited, structured reports	Web + email, Teams, meetings, files	Structured report with citations — optimized for multi-step tasks that typically take longer than a standard Chat response
Chat	Quick answers from live M365 data — email, calendar, Teams	Your live M365 data only	Conversational — fast
Analyst	Data analysis and calculation — Excel, CSV, uploaded files	Uploaded documents and files	Insights, charts, tables — analytical processing on uploaded files

How to access Researcher

Open Copilot

Go to m365.cloud.microsoft or open the Microsoft 365 Copilot app. Sign in with your M365 Copilot licensed account.

Select Researcher

Click the Agents button in the chat interface (bottom-left of the input area or sidebar). Select Researcher from the list of built-in agents.

Enter your research prompt

Type your research question in the chat input. Be specific about what you need — "Summarise the competitive landscape for X" or "What are the latest regulatory developments in Y". Researcher will run multiple steps before responding.

Review and refine

The output is a structured report with cited sources. You can ask follow-up questions, request a shorter version, or ask Researcher to focus on a specific sub-topic.

Export or share

Copy the output to a Copilot Page (shared editable doc) via the Create page option. Or copy and paste into Word, email, or your reporting tool.

Enable Frontier modes (if enrolled)

If your org is in the Frontier programme and your admin has enabled the Anthropic sub-processor, Critique mode (Roadmap 553213) runs automatically on each Researcher response. Council mode (Roadmap 558538) can be toggled on — it runs two AI models in parallel and compares their outputs. Both modes are controlled exclusively by the Anthropic sub-processor setting. If Anthropic is blocked by your admin, neither mode is available.

Frontier programme — Critique & Council modes

Available to orgs enrolled in Microsoft 365 Copilot Frontier

What Critique and Council modes do

Mode 1 · Roadmap 553213

Critique mode

A two-model pipeline: the first model drafts the research report; a second model reviews it as an expert and strengthens citation grounding, structure, and accuracy. Microsoft reports internal benchmark gains vs competing research tools (per Microsoft blog, April 2026). Enabled by default in Frontier — no extra step needed. Admin dependency (MC1265765): Critique has no separate admin control. It is enabled or disabled exclusively by the Anthropic sub-processor setting in the M365 admin centre. In EU/EFTA/UK tenants, Anthropic is off by default — admin opt-in required. In Government/Sovereign cloud, Anthropic is fully excluded and this mode is unavailable.

Frontier

Mode 2 · Roadmap 558538

Council mode

Multiple frontier models — which may include third-party providers listed as Microsoft sub-processors — run in parallel on the same research prompt. Each produces a standalone report. A judge model then compares them, highlighting where they agree, where they diverge, and what each caught that the other missed. Admin dependency (MC1265765): Council has no separate admin control. Like Critique, it is enabled or disabled exclusively by the Anthropic sub-processor setting in the M365 admin centre. Admins who enable Anthropic unlock both modes; admins who block Anthropic disable both. In EU/EFTA/UK tenants, Anthropic is off by default — admin opt-in required. In Government/Sovereign cloud, Anthropic is fully excluded and this mode is unavailable. Toggle on from the Researcher interface once Frontier enrolment and Anthropic sub-processor consent are confirmed.

Frontier

Use cases — what to ask and what to expect

Each card shows example prompts and who it is for

Use cases — by role and scenario

Competitive intelligence briefing

BD · Strategy · Executive

Most common use case

What Researcher searches

Web — news, press releases, reports
Your emails mentioning the company or topic
Your Teams conversations and meeting notes
Your OneDrive and SharePoint files

Example prompts

Give me a competitive briefing on [company] — recent news, products, and positioning.
What are the five biggest AI deployments in the EPC sector in the last 12 months?
Summarise everything I have on [topic] across my emails, Teams, and files from the last 6 months.

Researcher cites each source in the output. Check citations before sharing externally — web sources can include outdated or unverified content.

Regulatory and policy research

Legal · Compliance · HSE · Technical Authority

High value for regulated industries

What Researcher searches

Regulatory agency websites and official publications
Your internal policies and procedures
Your emails and Teams messages on the topic

Example prompts

What are the current EU AI Act obligations for AI systems used in safety-critical processes?
Summarise the latest OSHA guidance on [topic] and flag any changes in the last year.
What does our internal policy say about [process] and how does it compare to the current regulation?

Researcher can read and synthesise web-based regulation, but it cannot access paywalled legal databases. Always verify citations against the official source before use in a legal or compliance context.

Internal knowledge synthesis

Programme Director · PM · EA · Department Head

High value for long-running initiatives

What Researcher searches

Your email threads on the project or initiative
Meeting recordings and transcripts
Teams channel messages and shared files
SharePoint documents and decision logs

Example prompts

Summarise everything that has happened on [project] across my emails and Teams in the last 30 days.
What decisions were made in the last three steering committee meetings?
Who has been involved in [initiative] and what has each person contributed so far?

Researcher respects M365 permissions — it only returns information from files and messages you already have access to. It cannot surface content that is not shared with you.

What Researcher can access

Researcher combines web search with your M365 work data. It does not access data outside your M365 tenant, and it does not return information from files or conversations you do not have permission to see.

Web search — public pages, news, documentation

Outlook email and calendar

Teams messages and channel conversations

Meeting transcripts and recordings

OneDrive and SharePoint files

Paywalled databases — legal, academic

Critique mode — dual-model review pipeline

Council mode — parallel multi-model comparison

Available now (GA)

Frontier programme only

Not accessible

Known limitations and considerations

Researcher takes too long or times out. Researcher runs multiple steps before returning — typical queries take noticeably longer than a standard Chat response — plan for at least a minute on complex topics. If you need a quick answer, use Chat instead. Very broad or open-ended prompts increase processing time without necessarily improving quality.

The output cites a web source that is outdated or inaccurate. Web sources are pulled in real-time but not editorially verified. Researcher cites what it finds, not what is necessarily correct. Always check citations before sharing research externally or using it in a decision.

Internal emails or meetings are missing from the output. Researcher respects M365 permissions. If a colleague's files or chats are not shared with you, they will not appear. For older historical content, specify a date range explicitly in your prompt — Researcher prioritises recent data, so anchoring the timeframe improves recall on older items.

Critique or Council mode is not available in the Researcher interface. Both are Frontier-only features (Critique: Roadmap 553213, Council: Roadmap 558538) and require your organisation to be enrolled in the Frontier programme. Crucially, both modes are controlled exclusively by the Anthropic sub-processor setting in the M365 admin centre — there are no separate toggles for each mode (source: MC1265765). If your admin has blocked Anthropic, neither mode is accessible even if you are enrolled in Frontier. In EU/EFTA/UK tenants, Anthropic is off by default and requires explicit admin opt-in. In Government and Sovereign cloud tenants, Anthropic is fully excluded and both modes are permanently unavailable. Contact your M365 admin to verify Frontier enrolment and Anthropic sub-processor consent.

The report does not match what Chat told you. Researcher searches more broadly than Chat and may surface sources Chat did not use. Discrepancies between Researcher and Chat outputs are normal — they draw on different query strategies. Use Researcher's citations to verify.