Researcher is one of two flagship named agents Microsoft shipped in Copilot Chat. It synthesizes information from your work data — emails, SharePoint, Teams, OneDrive — alongside the web, and produces full reports. Think Deep Research, but with access to your company's context. The skill is knowing when that blend is its edge — and never trusting a number it pulls without checking.
Step 1 of 60% complete
The mental model
Deep research, plus your company's context
A public research tool can read the whole web but knows nothing about your business. A plain chatbot knows your files but won't go research the market. Researcher does both at once — and the combination is where it earns its keep.
Predict first
You ask for a competitive analysis on a rival. What can Researcher do that a public deep-research tool fundamentally can't?
The killer pattern
Ask a question that needs both sources
The best Researcher prompts deliberately span the outside world and your own data. Here's a competitive analysis for a customer meeting, split by where each piece comes from:
From the web
Their public positioning & pricing
Product changes in the last 6 months
Analyst coverage
Recent funding or layoffs
From your M365 data
Sales notes mentioning them (last 90 days)
Product marketing's battlecard
Customer feedback comparing you to them
Win/loss notes from deals
Then you ask for a structured output: their positioning vs yours, their top 3 strengths, your 3 angles, and 3 talking points — with every internal source cited by document name and date so you can verify.
Why cite internal sources
Citations by file name turn an impressive report into a checkable one. You can open the exact battlecard or sales note and confirm Researcher read it right.
Do it · verify the numbers
Catch the figure Researcher misread
Researcher drafted a slice of a QBR from your real data. It reads cleanly — which is exactly the danger. Below is the source data, and below that the draft. One figure in the draft doesn't match the source. Tap it.
Source · Q1-2026-Financials.xlsx
Q1 revenue
$4.2M
New customers
137
Churn rate
3.1%
NPS
52
Researcher draft · Q1 executive summary
Q1 revenue landed at , up 12% on plan. We added new customers, and churn improved to . Customer sentiment held strong, with an NPS of .
Tap the figure that doesn't match the source table.
More ways to use it
Two more patterns worth keeping
Long-form report from scattered sources
Point Researcher at five places — financials, pipeline, CS review, eng metrics, the strategy doc — and have it draft a structured QBR, even matching last quarter's voice from a prior file.
Decision-support, not a verdict
Ask for the 3 strongest reasons for, the 3 against, the one data point that should decide it, and 2 questions to ask first. Tell it not to give a yes/no — it'll respect that.
Decision supportShould we approve Marketing's conference sponsorship? Pull the request, our last 4 quarters of marketing-spend ROI, past sponsorship outcomes, and industry benchmarks. Don't tell me yes/no — give 3 reasons for, 3 against, the 1 data point that should drive it, and 2 questions to ask Marketing first.
Researcher only sees what your M365 permissions allow. A restricted SharePoint site is invisible to it — correct behavior, but it means a source can be silently missing from the report.
The boundary
When a source goes missing
Researcher's report on a rival is solid — except it never mentions the big deal your team lost to them last quarter, which everyone on the deal team knows about. What's the most likely reason?
🔎
Lesson complete
Researcher is your analyst — and you're its editor
What you can do now
Reach for Researcher when a question needs internal data + web context
Split a prompt explicitly into web sources and your-data sources
Make it cite internal sources by file name so you can verify
Check every number against the source — a misread cell flows through the whole report
Remember permission boundaries: a restricted site is silently invisible
Your move: run one hybrid Researcher task
Pick a real decision or report you owe someone this week. Frame it as a Researcher prompt that needs both your company's data and web context — and tell it to cite internal sources. Then verify the numbers before you send. Compare the time and quality to doing it by hand.
Agents · Lesson 02 — Analyst: turn data into insight
Analyst is Researcher's quantitative sibling. Where Researcher synthesizes documents and the web, Analyst crunches the numbers — Excel-heavy analysis, charts, and statistical insight, all from natural-language prompts.
Hey! I'm your AI Coach for this lesson on Copilot Researcher. Ask me how to frame a hybrid research prompt, when to use Researcher vs Analyst, or how to verify what it pulls. What are you researching?
Free lesson coaching is limited to 3 questions. Upgrade to Pro for unlimited coaching on every lesson.