The Copilot Researcher agent: what it does, the data it uses, and where it fails

Researcher is one of the built-in Microsoft 365 Copilot agents, and it is the one people reach for first. This page covers what it is built to do, the data it draws on, the failure modes to expect, and how it differs from its sibling, the Analyst agent. It is one of the two deep dives in our ranking; the shortlist lives at the best Copilot agents. If you need the product concept first, see what a Copilot agent is.

What Researcher is built for

Researcher is an advanced reasoning agent. The point of it is the task a single chat reply handles badly: gathering from several sources, weighing them, and returning a structured brief with its reasoning. Microsoft frames it as an agent that combines web data with your own work content, supports deeper reasoning, and produces structured output. The details are in Microsoft's Researcher agent documentation.

In practice, that means you give it a question that has parts — compare these options, assess this market, summarize what is known about this topic — rather than a single fact to look up. It plans, works through the steps, and writes up what it found.

The data it uses

Researcher works across two kinds of source:

• Public web sources. The open web, the same material any research assistant can reach.
• Your work content. Files, messages, and sites inside your organization — but only what you are permitted to see. It does not widen your access.

That second point is the one people trip over. If a document is not indexed, not in a readable format, or sits somewhere the agent cannot reach, it will not be part of the answer, and the agent may not tell you it was missing. Give it the material explicitly when the source matters.

A worked example

A task shape Researcher handles well is a decision brief. For example:

The instruction to cite sources is doing real work: it makes the output checkable, and it exposes when the agent is inferring rather than quoting. Ask for the structure you want — a table, a set of headings, a short brief — and Researcher is more likely to return something you can use.

Where it fails

• It is slow. It plans and runs multiple steps, so it is not the tool for a quick answer. That is the trade for depth.
• It may ask instead of answer. If the brief is ambiguous, it can come back with clarifying questions. That is better than guessing, but it means the first request often is not the last.
• It misses what it cannot reach. Unindexed or permission-blocked content simply is not in the answer, and the miss can be silent.
• It can be confidently wrong. A well-structured brief is still a draft. Conclusions and cited figures should be checked against the source before you rely on them.
• It is capped. Usage is limited on eligible plans, and the allowance is shared with the Analyst agent. The exact figure changes, so check your plan.
• It is plan-dependent. Whether you have Researcher at all depends on your Microsoft 365 licensing and your organization's configuration.

Researcher and Analyst

Researcher and Analyst are siblings, and they are easy to confuse. The short version: Researcher reads and reasons across sources; Analyst works with data and numbers. If your question is "what does the evidence say?", that is Researcher. If it is "what do these numbers show?", that is the Copilot Analyst agent. They share a usage allowance, which is one practical reason to pick the right one first.

Getting better results from Researcher

• Ask for the output format you want up front: brief, table, or headings.
• Name the sources or folders to use, rather than relying on discovery.
• Tell it to cite where each point came from.
• Split a sprawling question into two focused ones.
• Review the brief against the cited sources before you act on it.

TopCopilotAgents is independent and not affiliated with Microsoft. General information only; confirm current behavior and limits with Microsoft.