A recent explainx.ai guide compared the two and said it could not confirm one detail: whether the often-quoted 25-query monthly limit still applies. Microsoft's current documentation answers part of that, and the answer matters to anyone planning a rollout.
The short version
| Researcher | Analyst | |
|---|---|---|
| Core job | Multi-step research ending in a cited report | Data analysis, trends, forecasts and visuals |
| Typical input | A question plus the sources it should use | Spreadsheets, CSVs or other structured files |
| How you check it | Source citations | Python code you can inspect |
| Web search | User toggle in the input box | No user toggle; follows tenant policy |
| Documented limit | 25 queries per user per month (current Researcher FAQ) | No separate current figure in the documents reviewed |
Researcher: slow on purpose
Microsoft's current Learn page says Researcher handles complex, multi-step research and produces source-cited reports. It draws on the web and on work content you can access, including files, emails, meetings and chats. Microsoft says the slower pace is deliberate: the agent spends more time retrieving and analysing material so it can write a deeper analysis. Standard Copilot chat is tuned for quick jobs such as summarising an email or drafting a short reply.
Microsoft's own decision guide comes down to this:
- Use standard chat for a quick summary, a short reply, light brainstorming, or a back-and-forth task that doesn't need heavy analysis.
- Use Researcher when you need reasoning across several web and work sources and a report you can share, with headings, visuals and citations.
Microsoft's current Researcher FAQ adds some practical details:
- Response time: simple queries typically finish in under five minutes. Very complex ones can take 10 to 45 minutes.
- Data sources: Researcher uses Microsoft Graph, connectors and the Bing index for recent web data. Archived mail is used as backfill when the primary inbox doesn't have enough.
- Model choice: inside the Researcher agent, a model picker lets you choose the model and mode, and Claude can be enabled. If you call @Researcher from an ordinary Copilot chat, the picker is disabled and the model is set to auto.
- Gaps: Researcher can't take images as input. Microsoft lists PowerPoint and PDF output, memory and control over report length as "to be released soon."
- Fixed behaviour: you can't customise or extend Researcher through Copilot Studio. Users also can't disable, remove or unpin it.
Prompting Researcher properly
Microsoft gives three tips:
- Be specific. A detailed question produces a more focused, more reliable report.
- Define the scope. Say whether you want workplace resources, the web or both.
- Answer the clarifying questions. Microsoft says this step is how you steer the research and keep the report on target.
A good pattern covers the question, the sources, any date limits, and what you want back. For example: name the competitor topic, point to a specific internal document, exclude older material, and ask for a comparison table plus a list of open questions. Answering the clarifying questions properly is how you get a usable report on the first run.
Analyst: a data scientist that shows its code
Microsoft introduced Analyst in March 2025 as an agent that "thinks like a skilled data scientist." It reasons through problems step by step and can run Python for complex data queries, and you can watch the code run and check its work. Microsoft's launch examples included:
- combining several spreadsheets into a demand forecast
- visualising customer purchasing patterns
- projecting revenue
Northwestern University's IT knowledge base gives a plain description: a specialized, built-in agent in Microsoft 365 Copilot that helps users analyze, manipulate, and visualize data from files, such as Excel spreadsheets or CSV files. It also says that while standard Copilot handles general productivity, the Analyst agent is specialized for deep, multi-step data analysis, including coding (Python) to solve problems.
Getting started with Analyst
Microsoft Support documents these steps:
- Open Microsoft 365 Copilot.
- Under Agents in the left navigation pane, select Analyst.
- Enter a question about the data you want to analyze.
- Select the + icon and then select Attach content to supply the file(s) that have the data you want to analyze.
What success looks like: Analyst returns calculations, tables or charts with the Python behind them. Read that code before you trust the chart. A chart you can't trace back to its method shouldn't go into a board pack.
If Analyst isn't there: Microsoft says if the Analyst agent is not available, it may be because your administrator has not enabled it. The interface may also differ from the steps above. The current Support page warns that Home, the new starting point in Copilot where Chat and Cowork come together, is still rolling out, so what you see in the Copilot app might be different from what this article describes. Microsoft's admin documentation also says Researcher and Analyst are reachable from Tools in Copilot Chat, so check both places.
A caveat on models: Microsoft's 2025 materials said Analyst was built on OpenAI's o3-mini and Researcher on OpenAI's deep research model. Those are launch-era statements. Researcher now has its own model picker, so don't treat the 2025 model names as current.
File limits: neither the explainx.ai guide nor the official pages reviewed here list a file-size or format limit for Analyst. Test with your largest real workbook before you tell a team it will work.
The 25-query question, partly settled
Microsoft announced general availability on June 2, 2025, for everyone with a Microsoft 365 Copilot licence. That announcement said licensed users could run up to 25 combined queries per month across Researcher and Analyst. It also said Researcher supported 37 languages and Analyst eight. Ohio State University's IT department reported the same arrangement in July 2025: both agents are pre-pinned in all Microsoft 365 Copilot apps and allow up to 25 combined queries per month for holders of a full Copilot licence.
The explainx.ai guide said it found no current Microsoft page stating a cap. Microsoft's current Researcher FAQ does state one: a maximum of 25 queries per user per month. Two qualifications:
- That FAQ entry covers Researcher. The documents reviewed here don't confirm whether Analyst currently shares a combined pool or has its own allowance. The "combined" wording comes from the 2025 announcement.
- The language figures have moved. The FAQ now says Researcher supports "over 30 languages, the same as Microsoft Copilot," so the 37/8 split is a 2025 figure.
Practical advice: treat 25 Researcher runs per user per month as the documented baseline. Before you plan a heavy pilot, ask your Microsoft account team how Analyst usage is counted in your agreement.
Licensing also deserves care. Microsoft's Copilot overview separates Copilot Chat (Basic), Microsoft 365 Copilot (Basic) and Microsoft 365 Copilot (Premium), and it puts some experiences under usage-based billing. One third-party guide says Analyst also comes with certain consumer Microsoft 365 Premium or Pro subscriptions. That claim isn't in the official pages reviewed here, so confirm it before you rely on it.
The admin side: web search, DLP and visibility
Admins can block the agents, but users can't hide them. Microsoft's admin documentation calls Researcher and Analyst first-party experiences that run inside the Microsoft 365 commercial data processing boundary and inherit its security, privacy and compliance commitments. They stay available under Tools even when agents are disabled more broadly, and admins can block them specifically.
Web search follows tenant policy. Microsoft's web-search documentation says its rules apply to Researcher, Analyst and Cowork. Web search isn't required, but Microsoft recommends enabling it. Researcher has a user-facing Web search toggle; Analyst and Cowork don't. Admins control web search with the Allow web search in Copilot policy in Cloud Policy service. If an admin chooses Disabled in Microsoft Copilot Work mode; Enabled in Microsoft Copilot Web mode and Microsoft Copilot Chat, web search is switched off in Researcher and Cowork. The Researcher FAQ adds two points:
- You can't allow or block individual websites. The only control is the global switch.
- With web search off, Researcher relies on a model trained on web data up to 2024. Keep that in mind when it reports on recent market events.
DLP applies. The Researcher FAQ says the agent follows Data Loss Prevention policies the same way Copilot does. It also says Researcher isn't trained on tenant data and doesn't send customer content to OpenAI. Purview DLP can block external web search when a prompt contains sensitive information types. Ask whether that rule is active before anyone puts regulated data into a research prompt.
Usage visibility is limited. The FAQ says there is "no existing reporting tool" for Researcher and Analyst usage. In the eDiscovery section, though, it says admins can see usage metrics such as how many times Researcher is used, but not the conversation content. Researcher session content isn't directly available to admins or compliance tools by default. Given that inconsistency, test what your tenant's reports actually show before promising anyone a dashboard.
Permissions are a risk as well as a protection. Researcher respects the access a user already has. If that user can open files they shouldn't, Researcher can read them too. Microsoft's overview points to SharePoint Advanced Management and restricted content discovery as ways to reduce oversharing. Do that clean-up before a wide rollout, not after the first awkward report.
Risks worth naming
- A citation doesn't prove a claim. It only shows where the claim came from. Open the source behind any figure that will leave your team.
- Prompt injection. Researcher reads emails, files and web pages, and any of them could carry hidden instructions. Keep human approval on anything that leads to action.
- Python can be wrong too. Visible code makes Analyst auditable, but it won't catch a bad join or a quietly dropped row unless someone reads it. Ask Analyst to list any rows it excluded and why.
A pilot that actually tells you something
- Pick three research questions your team already answers by hand each month. Run each in Researcher with a defined scope. Record the time taken, the citations you checked and any errors.
- Pick two spreadsheet questions. Run each in Analyst, read the Python, and compare the result with a manual calculation.
- Count runs per user against the 25-query baseline.
- Confirm which web sources and data each pilot user can actually reach under tenant policy.
- Keep a use case only if the checked output beats the manual version, and save the prompts that worked in a shared library.
Bottom line: use Researcher for an evidence-backed brief and Analyst for numbers you can audit. Use standard chat for everything quick. Microsoft's current documentation now states the 25-query monthly limit for Researcher, so plan around it and confirm how Analyst is counted.
References
- Copilot Researcher vs Analyst: What Each Agent Does (2026) - explainx.ai explainx.ai · 2026-11-19T00:00:00+00:00
- Researcher and Analyst are now generally available | Microsoft Copilot Blog microsoft.com
- Introducing Researcher and Analyst in Microsoft 365 Copilot | Microsoft Copilot Blog microsoft.com