Generative AI in Your Research
Wednesday 4 November 2026 · 12:45–17:00 · lunch from 12:00
Agenda
Your team today
Where does AI actually help me? – how this tool works
We built this tool for the research-process session: so that the room decides together where AI genuinely helps your research – and where it does not – and everyone leaves with tested starting points.
Research value
Your group enters its discipline and goes through the phases of the research process from that point of view: how valuable for your research, what kind of gain, how feasible. Star what you want to try in Step 2, flag what is “not an AI task” or needs a rules check, and add missing use cases.
→ Results: the room’s value × feasibility matrix shows which use cases win.
Try it with Copilot
Each group picks one to three use cases and tests them live with Copilot – with a short guide, a starter prompt and the practice data. Record how it went and what you noticed, then share it with the room.
→ Round 1, exchange, round 2, exchange – then three findings and one open question per group.
Keep using it
The results report, the guides, the practice data, Tom’s slides and the overview of academic extensions and tools stay available – as a starting point for your own work.
You told us
What 37 of you said in the pre-workshop survey – and how it shaped this afternoon. Move through the slides with the arrow keys or the buttons; F opens full screen.
Where does AI really help your research?
Choose a phase, open a use case and rate it. Step 1 is about the value for your research; in Step 2 your group tries one to three use cases live with Copilot. Work from the point of view of your group’s discipline.
Practice data
Open papers and raw datasets on weather and climate change for Step 2 – use them instead of your own unpublished data. Everything here may be shared and uploaded to Copilot under the licence shown for each item.
How to use it with Copilot
- Download papers and data and unzip the pack – or download single files below.
- Pick a test idea below, or any paper or data file that fits your use case.
- In Copilot Chat, upload the paper or data file you need for your prompt.
- For a data file, first ask Copilot to describe it: columns, units, missing-value codes. Check this against the file.
- Then run your use case – and check every number, quote and citation against the data or the PDF.
Copilot reads a PDF as text, so tables, figures and equations are easily misread – check them in the paper itself. If Copilot does not accept a .txt file, rename it to .csv or paste its content into the chat.
Results from the room
Value against feasibility, top use cases, and what each group found.
Value × feasibility
Top use cases
By value × feasibility. Select one to open it.
Results report
Fills automatically – the basis for the follow-up e-mail.
Academic Extensions & Tools
Tools commonly used in academic work, mapped to the research phases – for orientation, not recommendations. They complement Copilot rather than replace it: subject databases, reference managers and analysis software do things a general assistant cannot. Many of them can also be connected to Copilot, most flexibly through the Model Context Protocol (MCP), an open standard that lets an assistant call another tool and use its results.
Example: a Copilot Studio agent connected to scite’s MCP server can check, while you draft a literature review, whether later studies support or contrast a finding you cite – and answers with the actual citation statements and DOIs instead of guessing.
Ratings: by students at a German university, 2024/25 (greyed out with fewer than three ratings); tools without a rating were added for this workshop. MCP official server from the provider · MCP community server, review before use · “Copilot Studio” = remote server that can be added there directly · Microsoft 365 other ways the tool connects to Copilot, Word, Teams, Outlook or Power Automate.
How to connect a tool to Copilot via MCPThree routes, from simple to flexible – and what to check first.
1 · Ready-made connectors
With a Microsoft 365 Copilot licence, IT can switch on ready-made Copilot connectors – for research, for example Consensus, scite, PubMed, DeepL and Notion. Many of them use MCP in the background. Nothing to build – ask your IT team whether they are available.
2 · Your own agent in Copilot Studio
- In Copilot Studio, create an agent and open Tools → Add a tool → New tool → Model Context Protocol.
- Enter a name, a short description (Copilot uses it to decide when to call the tool) and the server URL from the provider’s documentation.
- Choose how to sign in: none, API key or OAuth 2.0. With OAuth, each researcher uses their own account.
- Add instructions such as “cite DOIs, never invent sources”, test the agent, then publish it to Teams and Microsoft 365 Copilot and share it.
3 · Code and analysis
GitHub Copilot in VS Code connects to MCP servers directly – including local ones such as Zotero or jamovi.
Without MCP
Many tools already work inside Microsoft 365: reference managers cite while you write in Word (Zotero, EndNote, Mendeley, Citavi, Paperpile), note-takers join Teams meetings (Otter, Sally), and some offer Power Automate connectors. Filter by “Microsoft 365 integration” below.
Before you start
- Copilot Studio only connects to remote servers (HTTPS). Local community servers need VS Code or a desktop assistant.
- Building and publishing agents may need additional licences; IT usually approves agents and can restrict connectors.
- Your prompts and inputs go to the tool provider. Check your university’s data rules; never send unpublished or personal data to an unvetted server.
- Community servers are not maintained by the provider – review the code and permissions first.
About Copilot
Copilot is the assistant universities provide with data protection when you sign in with your university account. Other assistants can serve many of the same use cases – check what your university permits. Assessments as of October 2026; features change frequently.