A
Agent
A Copilot agent is an AI assistant with a defined task, its own instructions and access to specific sources or actions. Unlike a one-off chat, an agent works purposefully and repeatably.
Deep diveCopilot knowledge base
Copilot terms, clearly explained
The Copilot knowledge base explains the core terms of Microsoft Copilot in plain language: from agent and grounding to Microsoft Graph and sensitivity labels. One place to quickly grasp what a term means and what it refers to in your environment.
How to read this
Terms, in three layers
A–Z glossary
Copilot terms at a glance
Sixteen core terms, each in one or two sentences — with a link to the deep-dive page.
C
Copilot Chat
Copilot Chat is the conversational interface where you ask questions and get tasks done, with or without access to your work data. For many people it is the starting point with Copilot.
Deep dive C
Copilot Studio
Copilot Studio is the environment where you build and publish your own agents: with their own knowledge sources, actions and rules. It makes Copilot specific to your processes.
Deep dive D
DLP
DLP (Data Loss Prevention) is the set of rules that stop sensitive information being shared improperly or leaving the organisation. Alongside Copilot, DLP keeps control of what AI may do with data.
Deep dive G
Grounding
Grounding means Copilot bases its answer on real, current sources rather than the language model alone. That makes answers specific to your organisation and far easier to verify.
H
Hallucination
A hallucination is an answer that sounds plausible but is factually wrong or invented. Strong grounding and citations reduce the risk, but a human check is still needed.
M
Microsoft Graph
The Microsoft Graph is the gateway to your organisation data in Microsoft 365: emails, files, chats, calendar and people. Copilot uses the Graph to base answers on work that already exists.
Deep dive O
Oversharing
Oversharing is when files are shared too widely, so people can see more than intended. Copilot exposes existing oversharing, because it finds everything someone has access to.
Deep dive P
Prompt
A prompt is the instruction or question you give Copilot. The more concrete the prompt (role, context, desired format), the more useful the answer.
Deep dive R
RAG
RAG (Retrieval-Augmented Generation) is the technique that first retrieves relevant information and only then generates an answer. Copilot uses it to ground answers in your own sources.
Deep dive R
RBAC
RBAC (Role-Based Access Control) grants access based on someone’s role. Copilot respects these permissions: a user only sees through Copilot what they were already allowed to see.
Deep dive S
Semantic index
The semantic index is a meaning-based layer on top of your Microsoft 365 data, so Copilot finds content by meaning and not just exact keywords. That makes grounding more accurate.
S
Sensitivity label
A sensitivity label classifies and protects a document or message, for example as Confidential. Copilot inherits these protections, so sensitive content stays protected.
Deep dive T
Tenant
A tenant is your organisation’s isolated environment within Microsoft 365, with its own users, data and settings. Copilot works strictly within the boundaries of your tenant.
W
Work vs. web grounding
With work grounding Copilot draws on your own Microsoft 365 data via the Graph; with web grounding it uses public web information. The choice decides whether an answer reflects internal or public knowledge.
Z
Zero Trust
Zero Trust is the security principle that no user or device is trusted automatically; every request is verified. It is the foundation for safe Copilot use.
Deep diveOur conviction
Understand what you deploy, before you deploy it.
In practice
Turning terms into use
From your first question in Copilot Chat to your own agent in Studio — this is how the terms fit together.
Start with Copilot Chat
The conversational interface is the low-threshold starting point for daily tasks.
Understand your grounding
Know whether an answer comes from your own data or the open web.
Write sharp prompts
Add role, context and format: ask more concretely, get a more useful answer.
Check sensitive output
Hallucinations happen; for important decisions a human stays in the loop.
Respect access rights
Copilot only shows what someone was already allowed to see.
Build on in Studio
Create your own agents with knowledge sources and actions for your processes.
FAQ
Veelgestelde vragen
Short answers to the most common questions about Copilot terms.
What is a Copilot agent?
A Copilot agent is an AI assistant with a defined task, its own instructions and access to specific sources or actions. Unlike a one-off chat, an agent works purposefully and repeatably, for example for support or onboarding.
What does grounding mean in Copilot?
Grounding means Copilot bases its answer on real sources rather than the language model alone. With work grounding those are your own Microsoft 365 files via the Graph; with web grounding, public web information.
What is the Microsoft Graph?
The Microsoft Graph is the gateway to your organisation data in Microsoft 365: emails, files, chats, calendar and people. Copilot uses the Graph to base answers on work that already exists in your environment.
What is the semantic index?
The semantic index is a meaning-based layer over your Microsoft 365 data. Because of it Copilot finds content by meaning and not just exact keywords, which makes grounding on your own data more accurate.
What is RAG and does Copilot use it?
RAG (Retrieval-Augmented Generation) first retrieves relevant information and only then generates an answer. Copilot works this way, so answers are grounded in your sources rather than coming purely from the model.
Can Copilot invent mistakes (hallucinate)?
Yes. A hallucination is an answer that looks right but is wrong or invented. Strong grounding and citations reduce the risk, but a human check is always needed for important decisions.
Does Copilot see more than an employee may see?
No. Copilot respects existing access rights (RBAC): someone only sees through Copilot what they could already open. It does expose existing oversharing, because it finds everything someone has access to.
What is oversharing and why is it a risk?
Oversharing is when files have been shared too widely, so people see more than intended. Because Copilot finds everything someone can reach, latent oversharing surfaces faster. Our readiness scan maps this out.
What is the difference between Copilot Chat and Copilot Studio?
Copilot Chat is the conversational interface where you ask questions and get tasks done. Copilot Studio is the build environment where you create your own agents with their own knowledge sources, actions and rules for your processes.
How does a sensitivity label protect data in Copilot?
A sensitivity label classifies and secures a document, for example as Confidential. Copilot inherits that protection, so sensitive content stays protected in answers and shared output, alongside DLP rules that stop improper sharing.
How do I learn to work well with Copilot myself?
Start with clear, concrete prompts and practise on daily tasks. For structured learning, separate training material exists at copilotcursus.nl; we focus on advice and implementation, tying concepts to your practice in a working session.
From concept to practice
We translate these concepts to your environment — secure, well-grounded and with a concrete plan.