Adding Knowledge Sources

Overview

An agent built entirely on authored topics can only answer the questions you anticipated. If a user asks something you did not build a topic for, the agent has nothing to offer. But your organization already has answers to thousands of questions, spread across SharePoint sites, policy documents, FAQs, and internal web pages.

Knowledge sources bridge this gap. They connect your Copilot Studio agent to your organization’s existing content, allowing the agent to generate accurate, grounded answers from documents and sites your team already maintains. Instead of scripting every possible answer, you point your agent at the content and let it do the work.

This video shows you how to add and manage knowledge sources in Copilot Studio within government cloud environments.

What You’ll Learn

• Knowledge source types: SharePoint, websites, documents, and more
• SharePoint integration: How to connect sites and respect permissions
• Document uploads: When and how to add standalone files
• Ongoing management: How to keep knowledge sources accurate and current
• Government compliance: Data residency, sensitivity labels, and permission models

Script

Hook: Make your agent an expert

Your agent knows exactly what you have told it through authored topics, and nothing more. If a user asks about your agency’s telework policy and you have not built a topic for that, the agent cannot help. But your agency probably has a twenty-page telework policy document sitting on a SharePoint site right now.

Knowledge sources change the equation. They let your agent draw from your organization’s actual content, from SharePoint sites, web pages, and uploaded documents, to answer questions dynamically. Instead of scripting every possible answer, you connect the agent to the content that already exists.

In the next ten minutes, you will learn how to connect your agent to the content that matters.

Types of knowledge sources

Copilot Studio supports several types of knowledge sources. The most common are SharePoint and OneDrive, which let your agent search and reference content stored on SharePoint sites and OneDrive locations within your tenant. Public websites and web pages can be added so your agent references external content, such as official government resource pages. Uploaded files including Word documents, PDFs, and text files can be added directly to the agent. And for more advanced scenarios, Dataverse tables and custom data through connectors are also available.

Here is how knowledge sources work in practice. When a user asks a question and the agent does not have a matching authored topic, it searches the connected knowledge sources. It finds relevant content, uses generative AI to synthesize an answer from that content, and returns a response grounded in your organization’s actual documents. The agent cites its sources so users can verify the information.

This raises an important question: when should you use knowledge sources versus authored topics? Use authored topics for structured, repeatable interactions where you want precise control over the conversation flow. For example, a password reset process should be an authored topic with specific steps. Use knowledge sources for broad question-and-answer scenarios where users might ask dozens of variations of questions that your existing documentation already covers. A benefits FAQ or a policy reference library is perfect for knowledge sources.

For government environments, the key consideration is that your knowledge sources must stay within compliance boundaries. Content in your government cloud SharePoint remains in your government cloud when the agent references it.

Adding SharePoint and web content

SharePoint is the most powerful knowledge source for government agents because most agencies already have their documentation there. To add a SharePoint site, navigate to the Knowledge section in your agent’s settings. Click “Add knowledge” and select SharePoint. Enter the URL of the SharePoint site you want to connect. The agent indexes the content on that site and makes it available for answering questions.

An important detail: the agent respects SharePoint permissions. If a user interacting with your agent does not have access to a particular document or site, the agent will not surface that content to them. This means your existing information governance model carries forward into your agent automatically.

Best practices for SharePoint knowledge sources: point to well-organized, up-to-date sites. If a SharePoint site has outdated policies mixed in with current ones, the agent may surface outdated information. Use specific site URLs rather than broad root sites. Adding your entire organization’s SharePoint root will include far too much content, leading to less relevant answers. Instead, add the specific HR policy site, the IT knowledge base site, or the facilities FAQ site.

You can also add public websites as knowledge sources. This is useful for linking to official government resources like agency websites or published regulations. Add the specific page URLs that contain relevant content.

For government environments, ensure that SharePoint sites are within your government cloud tenant. External URLs should point to trusted resources, ideally .gov or .mil domains. Before connecting any knowledge source, verify that the content is approved for the audience your agent serves. If your agent is available to contractors, make sure the knowledge sources do not include employee-only content that contractors should not access.

Here is an example. You are building an HR support agent. Add your agency’s intranet HR policy site, which contains leave policies, benefits guides, and onboarding documents. Users can now ask “What is the telework policy?” or “How do I enroll in the health plan?” and the agent generates answers from the actual policy documents on that site.

Uploading documents

Sometimes the content your agent needs is not on a SharePoint site. It exists as standalone documents: a travel policy manual, a standard operating procedure, or a training guide. For these cases, upload the documents directly to your agent.

To upload a file, go to the Knowledge section and click “Add knowledge,” then select “Files.” Upload your document. Copilot Studio supports Word documents, PDFs, and text files. The agent processes and indexes the content, making it available for answering questions.

Document upload is the right choice when content exists in standalone files that are not published to a SharePoint site, when you have standard operating procedures or policy manuals that users frequently reference, when the content is relatively stable and does not change frequently, or when you want tight control over exactly which documents the agent references.

To get the best results from uploaded documents, use well-structured files with clear headings and sections. The agent uses document structure to understand context and provide relevant answers. Keep individual documents focused on a single topic area rather than uploading one massive document that covers everything. Update documents when the source content changes, and remove outdated documents to prevent the agent from surfacing stale information.

Be aware of file size and upload limits for your environment. If you have large documents, consider breaking them into smaller, focused files. A fifty-page policy manual could be split into separate files for each major section: travel authorization, reimbursement, per diem rates, and exceptions.

Here is a government scenario. You are building an employee services agent. Upload the agency’s travel policy manual, the telework agreement template, and the benefits enrollment guide. Users can ask specific questions like “What is the per diem rate for Washington DC?” or “When is open enrollment?” and the agent finds the answer in the uploaded documents.

Managing and updating knowledge

Connecting knowledge sources is not a one-time task. The content your agent draws from needs ongoing management to ensure answers remain accurate and current.

Start by reviewing knowledge source performance. Check which sources are being used most frequently. Identify sources that users query but that do not return useful answers, which may indicate the content is not well-suited for the questions being asked. Remove sources that are outdated or no longer relevant.

Different knowledge source types have different update behaviors. SharePoint sources update automatically as the site content changes. When someone publishes an updated policy on the SharePoint site, the agent’s knowledge reflects that change. Uploaded documents, however, must be manually re-uploaded when the source content is updated. If you upload a travel policy in January and the policy changes in March, you need to re-upload the updated document. Websites are re-indexed periodically, but the timing depends on your configuration.

Content quality directly impacts answer quality. If your knowledge sources contain conflicting information, the agent may give inconsistent answers. If documents are poorly written or disorganized, the agent struggles to extract relevant content. Review the actual content your knowledge sources point to and ensure it is accurate, well-organized, and consistent.

Create a management plan for your knowledge sources. Group related sources logically and use descriptions to document what each source covers. Create a refresh schedule for uploaded documents, checking quarterly at minimum whether documents need updating. Assign an owner for each knowledge source who is responsible for keeping it current.

For government compliance, audit which content is accessible through your agent. Ensure no classified or restricted content is connected to agents serving unclassified environments. Maintain records of knowledge source changes for compliance reviews and audits.

Government cloud considerations

Government cloud environments add specific requirements for knowledge source management that commercial environments do not face.

Data residency is the foundational concern. Knowledge sources in GCC, GCC High, and DoD environments stay within their respective cloud boundaries. When your agent searches a SharePoint site in GCC High, that search and the resulting data never leave the GCC High boundary. This is enforced by the platform, not something you need to configure, but it is important to understand and communicate to stakeholders who ask about data handling.

SharePoint permissions provide a critical layer of security. The agent can only surface content that the current user has access to. If a SharePoint site has different permission levels for different user groups, the agent respects those levels automatically. This means you can connect a broadly scoped knowledge source and still maintain access control. Do not use the agent as a way to bypass information governance. The permissions model exists for a reason.

Sensitivity labels from Microsoft Purview add another layer. Content labeled with sensitivity labels for restricted access will not be surfaced to unauthorized users through the agent. If your agency uses sensitivity labels to classify documents, those classifications carry forward into agent interactions.

From a compliance standpoint, document what knowledge sources are connected to each agent. Include knowledge source management in your overall agent governance plan. Schedule regular audits of connected content to verify that nothing inappropriate has been added and that outdated content has been removed.

One critical technical point: knowledge sources must be in the same cloud environment as the agent. You cannot connect a GCC High agent to a GCC SharePoint site or vice versa. If your agency operates across multiple cloud environments, each environment’s agents must use knowledge sources from within that same environment.

Close: Knowledge is power

Let us recap. Knowledge sources transform your agent from a scripted Q&A tool into an intelligent assistant that can draw from your organization’s actual content. SharePoint sites provide automatic updates and permission-based access. Uploaded documents work for standalone policies and procedures. And websites let you reference external resources.

Ongoing management is essential. Knowledge sources are only as good as the content behind them. Regular reviews, updates, and audits keep your agent’s answers accurate and compliant.

Here are your next steps. Identify the top content sources your users need access to, whether that is an HR policy site, an IT knowledge base, or a set of standard operating procedures. Connect a SharePoint site as your first knowledge source and test the answers your agent generates. Check whether the answers are accurate, relevant, and grounded in the source content.

An agent with knowledge sources does not just answer questions. It makes your organization’s content accessible and actionable for every user who needs it.

Sources & References

• Knowledge sources in Copilot Studio — Official documentation for knowledge source types, configuration, and management in Copilot Studio
• Generative answers from SharePoint and OneDrive — Guide to using SharePoint and OneDrive content for generative answers in agents
• Microsoft Copilot Studio documentation — Comprehensive documentation hub for all Copilot Studio capabilities