This chapter covers Microsoft Copilot for Microsoft 365, an AI-powered productivity tool that integrates with Microsoft 365 apps to assist users in creating, summarizing, analyzing, and automating tasks. For the MS-900 exam, this topic appears in approximately 10-15% of questions, primarily under Objective 2.4 (Describe Microsoft 365 productivity solutions). You will need to understand Copilot's capabilities, how it works, its licensing requirements, data privacy and security features, and how it differs from other AI offerings like Bing Chat Enterprise and Copilot for Security. This chapter provides a detailed, exam-focused breakdown of all these aspects.
Jump to a section
Imagine you are a busy executive with a highly skilled personal assistant. You don't just give vague instructions; you say, 'Get me the Q3 sales numbers from the Contoso account, summarize the key trends in a paragraph, and draft an email to the VP explaining the drop in renewals.' The assistant knows where all your files are (OneDrive, SharePoint, email, CRM), understands context (who the VP is, previous emails), and uses a set of tools (Copilot skills) to retrieve, analyze, and generate content. But crucially, the assistant never learns your secrets; after each task, they forget everything except the final output you approved. They don't store your data, and they cannot act on your behalf without your explicit instruction. Microsoft Copilot for Microsoft 365 works exactly like this: it's an AI assistant deeply integrated into the Microsoft 365 ecosystem, with access to your Graph data (emails, files, chats, meetings) but bound by strict security and privacy boundaries. It processes natural language prompts, uses the Microsoft Graph to fetch relevant information, and generates responses using large language models—all while inheriting your existing permissions and never persisting your data or training the underlying models on it.
What is Microsoft Copilot for Microsoft 365?
Microsoft Copilot for Microsoft 365 is an AI assistant that combines large language models (LLMs) with your organizational data from the Microsoft Graph and the Microsoft 365 apps you use daily. It is not a standalone product but an integrated experience within Word, Excel, PowerPoint, Outlook, Teams, and other M365 apps. Copilot helps you create documents, analyze data, summarize meetings, draft emails, and more, all while respecting your permissions and data security.
How It Works Internally: The Orchestration Flow
When a user submits a prompt (e.g., "Summarize the key decisions from yesterday's project review meeting"), Copilot follows a multi-step orchestration process:
Prompt Pre-processing: The user's natural language prompt is sent to the Copilot system. The system first checks the user's identity and permissions via Azure Active Directory (now Microsoft Entra ID). It also determines which Microsoft 365 service (e.g., Teams, Word) is the context.
Grounding via Microsoft Graph: Copilot queries the Microsoft Graph to retrieve relevant data. This includes emails, calendar events, chat messages, documents, and meeting transcripts. Crucially, Copilot only accesses data the user already has permission to see. If the user does not have read access to a file, Copilot cannot retrieve it.
Prompt Engineering and Grounding: The original user prompt is combined with the retrieved data to form a "grounded" prompt. This prompt includes the retrieved context (e.g., meeting transcript) and instructions for the LLM. The LLM is not directly exposed to the internet; it only receives this grounded prompt.
LLM Processing: The grounded prompt is sent to a large language model hosted in Microsoft's Azure OpenAI service. The model generates a response based on the provided context. The model does not retain the prompt or response after processing.
Post-processing and Compliance: The generated response is checked for policy compliance (e.g., sensitive data leakage prevention, DLP policies). It is also post-processed to format it appropriately for the target app (e.g., inserting into a Word document, displaying in a Teams chat).
Response Delivery: The final response is delivered to the user within the Microsoft 365 app. Citations are provided where possible, showing which source documents were used.
Key Components and Defaults
Copilot License: Requires a Microsoft 365 E3, E5, Business Standard, or Business Premium subscription (for desktop apps) plus a Microsoft Copilot for Microsoft 365 add-on license (currently $30/user/month).
Microsoft Graph: The data source for grounding. It indexes all user content in Exchange Online, SharePoint Online, OneDrive for Business, Teams, and other M365 services.
Large Language Model: Uses GPT-4 (or later versions) from OpenAI, deployed in Azure's infrastructure. The model is stateless; no data is used for retraining.
Permissions: Copilot inherits the user's existing permissions via Azure AD/Entra ID. It cannot elevate privileges.
Data Residency: Data processing occurs in the same geography as the tenant's Microsoft 365 data, unless configured otherwise.
Audit Logging: All Copilot interactions are logged in the Microsoft 365 Purview compliance portal for audit and eDiscovery.
Default Retention: Prompts and responses are not stored by Copilot beyond the immediate session. Microsoft does not use customer data for model training.
Configuration and Management
Administrators can manage Copilot via the Microsoft 365 admin center and the Copilot page. Key settings include:
Enable/Disable Copilot: Turn on or off for specific users or groups.
Data Access Controls: Configure which data sources Copilot can access (e.g., only SharePoint, not Exchange).
DLP Integration: Apply data loss prevention policies to Copilot responses.
Audit Logging: Enable detailed logging of Copilot interactions.
Power Platform Integration: Extend Copilot with custom connectors and plugins via Microsoft Copilot Studio (formerly Power Virtual Agents).
Interaction with Related Technologies
Copilot for Microsoft 365 vs. Bing Chat Enterprise: Bing Chat Enterprise is a separate service that provides AI chat with web grounding, using public web data. It does not have access to your Microsoft Graph data. Copilot for Microsoft 365 is grounded in your organization's data.
Copilot for Security: A separate product focused on security analysts, using threat intelligence and security data.
Microsoft 365 Copilot vs. GitHub Copilot: GitHub Copilot is for code generation in IDEs; it is a different product.
Copilot in Power Platform: Allows building custom AI-powered copilots using Microsoft Copilot Studio, which can be integrated with M365 Copilot.
Exam-Relevant Details
Licensing: You need a qualifying M365 plan (E3, E5, Business Standard, Business Premium) plus the Copilot add-on. E3/E5 are required for full desktop app integration; Business Standard includes web and mobile versions.
Data Privacy: Microsoft guarantees that your data is not used to train the AI models. This is a key differentiator from public AI services.
Grounding: The process of retrieving relevant data from Microsoft Graph to provide context to the LLM.
Copilot Studio: A low-code tool to create custom copilots and plugins.
Limitations: Copilot may make mistakes (hallucinations). It cannot perform actions on behalf of the user without explicit confirmation (e.g., sending an email requires user approval).
Common Exam Traps
Trap 1: Confusing Copilot for Microsoft 365 with Bing Chat Enterprise. Bing Chat Enterprise uses web data; Copilot uses your M365 data.
Trap 2: Thinking Copilot requires a separate subscription without a base M365 plan. It's an add-on.
Trap 3: Believing Copilot can access data the user doesn't have permission to. It inherits permissions.
Trap 4: Assuming Copilot stores prompts and responses for training. It does not; data is not retained beyond the session.
Verification Commands (for admins)
To check Copilot licensing via PowerShell:
Get-MgUserLicenseDetail -UserId user@contoso.com | Select-Object SkuPartNumberLook for COPILOT_M365 in the SkuPartNumber.
To enable Copilot for a user via admin center: Go to Users > Active Users > select user > Licenses and Apps > assign Copilot license.
User Submits a Prompt
The user types or speaks a natural language prompt in a Microsoft 365 app (e.g., 'Draft a follow-up email to the client about the proposal'). The app sends this prompt to the Copilot backend. The system first authenticates the user via Azure AD/Entra ID and identifies the app context (e.g., Outlook). It also checks the user's Copilot license entitlement. If the user is not licensed or the feature is disabled, the request is rejected. The prompt is then pre-processed to extract intent and entities.
Grounding via Microsoft Graph
Copilot queries the Microsoft Graph to retrieve relevant data based on the prompt. For example, if the prompt asks about a meeting, Copilot fetches the meeting transcript, calendar item, and related emails. The Graph query respects the user's permissions and returns only data the user can access. This step ensures the LLM has context from the user's own work. The retrieved data may include text from documents, chat history, email threads, or SharePoint pages.
Prompt Engineering and Grounding
The original user prompt and the retrieved Graph data are combined into a single grounded prompt. This prompt includes instructions for the LLM on how to use the context (e.g., 'Using the following meeting transcript, summarize key decisions'). The grounded prompt is then sent to the LLM. This step prevents the LLM from generating responses based on its training data alone; it must use the provided context.
LLM Generates Response
The grounded prompt is processed by a large language model (GPT-4) hosted in Azure OpenAI. The model generates a response token by token, based on the context. The model does not retain the prompt or response after generation. The response is then sent back to the Copilot backend. The entire process is stateless; no customer data is used for model improvement.
Post-processing and Compliance Check
The generated response is checked against organizational policies. This includes data loss prevention (DLP) rules, sensitivity labels, and compliance policies. If the response contains sensitive information (e.g., credit card numbers), it may be blocked or redacted. Citations are added to indicate which source documents were used. The response is also formatted for the target app (e.g., HTML for email, plain text for Teams).
Response Delivered to User
The final response is displayed to the user in the Microsoft 365 app. The user can review, edit, or discard the response. If the user approves, Copilot may perform actions (e.g., send email, insert into document). All interactions are logged in the Microsoft 365 audit log for compliance and eDiscovery purposes. The user can also provide feedback (thumbs up/down) to improve the system.
Enterprise Scenario 1: Automating Meeting Follow-ups
A large consulting firm uses Copilot to automate meeting follow-ups. After a Teams meeting, a partner asks Copilot in the meeting chat: 'Summarize the action items and assign them to team members.' Copilot retrieves the meeting transcript, identifies action items, and drafts a summary. It then creates a Word document with the summary and emails it to participants. The firm configured Copilot to always include a disclaimer and to respect sensitivity labels. The main challenge was ensuring that Copilot correctly identified action items from ambiguous conversation. They mitigated this by training users to clearly state action items and by using Copilot's 'customize' feature to add specific instructions.
Enterprise Scenario 2: Drafting Proposals from CRM Data
A technology company integrated Copilot with their Dynamics 365 CRM. When a sales rep wants to draft a proposal, they prompt Copilot in Word: 'Create a proposal for Contoso using the latest pricing and case studies from our SharePoint.' Copilot queries Microsoft Graph to pull relevant documents from SharePoint and OneDrive, and also accesses Dynamics 365 data via a custom Copilot plugin. The draft is generated with placeholders for specific details. The company had to carefully manage permissions to ensure reps only accessed data they owned. They also set up DLP policies to prevent copying sensitive pricing into unapproved documents.
Enterprise Scenario 3: Employee Onboarding Assistant
A multinational corporation built a custom onboarding copilot using Copilot Studio. New hires ask questions like 'What is the IT policy for remote access?' or 'Who is my onboarding buddy?' The copilot retrieves answers from HR SharePoint sites, FAQs, and employee directories. It is integrated with M365 Copilot so that when a manager says 'Prepare onboarding plan for new hire John,' Copilot generates a checklist and schedules meetings. The company faced performance issues when the copilot was initially deployed without caching, causing slow responses. They optimized by pre-loading frequently accessed data and setting response timeouts.
Common Misconfiguration Issues
Overly broad permissions: If users have too much access, Copilot may retrieve sensitive data unintentionally.
Lack of DLP policies: Without DLP, Copilot could generate responses containing confidential information.
Insufficient training: Users may not understand Copilot's limitations and may blindly trust outputs, leading to errors.
Ignoring audit logs: Without monitoring, organizations may miss inappropriate usage or data leaks.
What MS-900 Tests on This Topic
MS-900 Objective 2.4 focuses on describing Microsoft 365 productivity solutions, including Copilot. Specifically, you need to know:
The purpose and capabilities of Microsoft Copilot for Microsoft 365.
Licensing requirements (base M365 plan + Copilot add-on).
Data privacy and security: Copilot uses your data only to generate responses; it does not train models on your data.
How Copilot differs from Bing Chat Enterprise (web grounding vs. organizational data grounding).
Integration with Microsoft Graph.
The role of Copilot Studio in extending Copilot.
Common Wrong Answers and Why Candidates Choose Them
Wrong Answer: 'Copilot for Microsoft 365 is free with any Microsoft 365 subscription.' Why chosen: Candidates confuse general AI features (like Editor) with Copilot. Reality: Copilot requires an additional license ($30/user/month).
Wrong Answer: 'Copilot can access any data in the organization regardless of user permissions.' Why chosen: Candidates think AI has 'super' access. Reality: Copilot inherits user permissions; it cannot access data the user cannot.
Wrong Answer: 'Copilot stores all prompts and responses to improve the AI model.' Why chosen: Common with public AI services. Reality: Microsoft does not use customer data for training; data is not retained.
Wrong Answer: 'Bing Chat Enterprise and Copilot for Microsoft 365 are the same thing.' Why chosen: Both use AI and Microsoft branding. Reality: Bing Chat Enterprise uses web data; Copilot uses M365 data.
Specific Numbers and Terms
License cost: $30/user/month (as of 2025).
Base plans: Microsoft 365 E3, E5, Business Standard, Business Premium.
Data source: Microsoft Graph.
LLM: GPT-4 (Azure OpenAI).
Customization tool: Microsoft Copilot Studio.
Compliance: Audit logging in Purview.
Edge Cases and Exceptions
If a user has a Copilot license but the organization has disabled Copilot for their group, they cannot use it.
Copilot in Word and PowerPoint requires desktop apps (included in E3/E5). Business Standard users get Copilot only in web and mobile versions.
Copilot cannot access on-premises data unless it is synced to Microsoft Graph (e.g., via Graph connectors).
Copilot responses may be inaccurate (hallucination); users must verify critical information.
How to Eliminate Wrong Answers
If an answer says 'Copilot trains AI models on your data,' eliminate it because Microsoft explicitly states it does not.
If an answer implies Copilot is included in all M365 plans, eliminate it because it requires an add-on.
If an answer says Copilot uses web data, eliminate it (that's Bing Chat Enterprise).
If an answer says Copilot can bypass permissions, eliminate it because it inherits user permissions.
Microsoft Copilot for Microsoft 365 is an AI assistant integrated into M365 apps, using Microsoft Graph to ground responses in your organizational data.
Copilot requires a qualifying M365 base plan (E3, E5, Business Standard, or Business Premium) plus an add-on license costing $30/user/month.
Copilot inherits user permissions; it cannot access data the user does not have permission to see.
Microsoft does not use customer data to train AI models; prompts and responses are not retained.
Copilot Studio allows you to build custom copilots and extend Copilot with custom plugins.
Copilot differs from Bing Chat Enterprise, which uses public web data and is included with M365 E3/E5/Business plans.
All Copilot interactions are logged in the Microsoft 365 audit log for compliance and eDiscovery.
Copilot can generate inaccurate responses (hallucinations); users should verify critical information.
These come up on the exam all the time. Here's how to tell them apart.
Microsoft Copilot for Microsoft 365
Grounded in organizational data via Microsoft Graph
Requires M365 base plan + Copilot add-on ($30/user/month)
Integrated into M365 apps (Word, Excel, Outlook, Teams, etc.)
Data privacy: no training on customer data, data stays within tenant
Can be extended with custom plugins via Copilot Studio
Bing Chat Enterprise
Grounded in public web data
Included with Microsoft 365 E3, E5, Business Standard, and Business Premium (no add-on)
Available as a standalone chat interface (bing.com/chat) and in Edge sidebar
Data privacy: no training on customer data, but uses web search
Limited extensibility; no custom plugin support
Mistake
Copilot for Microsoft 365 is the same as Bing Chat Enterprise.
Correct
Copilot for Microsoft 365 is grounded in your organizational data via Microsoft Graph, while Bing Chat Enterprise uses public web data. They are separate products with different capabilities and licensing.
Mistake
Copilot stores all my prompts and responses to improve the AI.
Correct
Microsoft does not use customer data to train its AI models. Prompts and responses are processed in real-time and are not retained or used for model improvement.
Mistake
Copilot can access any data in my organization, even if I don't have permission.
Correct
Copilot inherits the user's existing permissions from Azure AD/Entra ID. It can only access data the user already has permission to view or edit.
Mistake
Copilot is a free feature included with all Microsoft 365 subscriptions.
Correct
Copilot requires a separate add-on license ($30/user/month) on top of a qualifying Microsoft 365 base plan (E3, E5, Business Standard, or Business Premium).
Mistake
Copilot can automatically send emails or make changes without user approval.
Correct
Copilot requires user confirmation before performing actions like sending an email or updating a document. It acts as an assistant, not an autonomous agent.
Reveal each answer, then mark whether you got it right. Score 60%+ to unlock the next chapter.
Yes. You need a qualifying Microsoft 365 base plan (E3, E5, Business Standard, or Business Premium) and then you must purchase the Microsoft Copilot for Microsoft 365 add-on license, which costs $30 per user per month. Without the add-on, you cannot use Copilot features even if you have a base plan.
Not directly. Copilot only accesses data that is indexed in the Microsoft Graph. If your on-premises data is synced to SharePoint Online or OneDrive for Business (e.g., via migration or using Graph connectors), then Copilot can access it. Otherwise, it cannot. For exam purposes, remember that Copilot relies on Microsoft Graph.
Yes, your data is safe. Microsoft explicitly states that customer data is not used to train the underlying AI models. Copilot processes your prompts and data in real-time to generate responses, but the data is not retained or used for model improvement. This is a key differentiator from public AI services.
Copilot in Teams can summarize meetings, suggest action items, and draft chat replies based on meeting transcripts. Copilot in Word helps you write, edit, and summarize documents. Both use the same underlying technology, but they are tailored to the context of the app. The exam may ask about specific capabilities in each app.
Yes, you can use Microsoft Copilot Studio (formerly Power Virtual Agents) to create custom copilots. These can be integrated with Microsoft Copilot for Microsoft 365 to extend its capabilities. For example, you can build a copilot that queries your CRM system and make it available within Teams.
The feature will be unavailable. The Copilot icon may be grayed out or hidden. The user will not be able to submit prompts. Administrators can assign Copilot licenses to users via the Microsoft 365 admin center.
No. Microsoft 365 Business Basic does not include the desktop versions of Office apps, and Copilot in desktop apps requires E3/E5 or Business Standard/Business Premium. However, Copilot in web apps may be available with Business Basic? Actually, as of the latest documentation, Business Basic does not qualify for Copilot. The qualifying plans are E3, E5, Business Standard, and Business Premium. So Business Basic is not supported.
You've just covered Microsoft Copilot for Microsoft 365 — now see how well it sticks with free MS-900 practice questions. Full explanations included, no account needed.
Done with this chapter?