Knowledge Sources Overview in Copilot Studio | Day 66

Day 66 of 365 Days of Copilot Studio gives a practical knowledge sources overview: where your agent gets information, how that information becomes grounded answers, and why source quality matters before you publish.

Day 65 focused on user permissions. Day 66 moves into the content layer. A Copilot Studio agent is useful only when it can find the right knowledge, respect access rules, and turn trusted information into answers users can act on.

This lesson also builds on Day 11: Mastering Knowledge Sources. Day 11 introduced the concept. Day 66 looks at knowledge sources from an architecture and governance point of view.

Note: Knowledge does not replace conversation design. It gives the agent the context it needs so answers are relevant, current, and grounded in your business information.

What Knowledge Sources Do

Knowledge sources give your agent information it can use to answer questions. Instead of relying only on a generic language model response, the agent can search trusted content, retrieve relevant information, and generate an answer grounded in that content.

The basic flow is:

User question -> knowledge retrieval -> grounded answer

That grounding is what turns a general assistant into a business-aware agent. The agent can answer using policies, procedures, documentation, structured business data, and other trusted sources.

An Agent Needs Context

Without knowledge, an agent may respond with a generic answer. It can explain concepts, but it may not know your organization's leave policy, support process, product guidance, or internal procedure.

With knowledge, the agent can answer in context. For example, if a user asks, "What is our leave policy?", a generic response may describe what leave policies usually contain. A grounded response can point to your actual policy, the number of allowed days, and where the user can find the full handbook.

Knowledge makes agents useful because it connects natural language questions to business-specific answers.

How Knowledge Reaches the User

When a user asks a question, Copilot Studio understands the request, searches connected knowledge sources, finds relevant information, and produces a response for the user.

The path looks like this:

1. User asks: the conversation starts with a business question.
2. Agent understands: Copilot Studio interprets what the user wants.
3. Knowledge is searched: relevant websites, SharePoint content, Dataverse data, files, or enterprise sources are considered.
4. Answer is grounded: the final response is based on retrieved information, not guesswork.

Question in, relevant knowledge out. That is the heart of knowledge-powered agents.

Connect Different Sources

One agent can use multiple knowledge sources. That matters because business information rarely lives in one clean place.

Knowledge source	Best fit	Example
Websites	Public or controlled web content	Product documentation, public FAQs, support pages.
SharePoint	Organizational documents and pages	Policies, procedures, intranet content, internal knowledge.
Dataverse	Structured business data	Cases, assets, requests, customers, approvals, reference data.
Files	Document-based knowledge	PDFs, Word documents, Excel files, handbooks, guides.
Enterprise data	Business systems and managed sources	Approved internal repositories and connected enterprise content.

The goal is not to connect everything. The goal is to connect the right sources for the questions the agent must answer.

Ground Answers in Web Content

Website knowledge is useful when trusted information already exists on a public or internal web page. It works well for documentation, FAQs, help content, and public information that changes over time.

Use website knowledge when:

• The content is already published and maintained.
• The page has clear headings and readable structure.
• The information is meant to be used by the agent's audience.
• The source owner has a process to keep the content current.

Trusted web content can turn static pages into conversational answers, but the page still needs content discipline. If the website is outdated, vague, or contradictory, the agent inherits that weakness.

Bring Organizational Knowledge

SharePoint is often where organizational knowledge lives: policies, procedures, internal guidance, project files, templates, and department documentation.

SharePoint knowledge is especially useful for employee-facing agents. An HR assistant, IT support agent, finance policy agent, or operations assistant can retrieve approved internal content and answer users without forcing them to search libraries manually.

For SharePoint knowledge, pay attention to:

• Structure: clear document titles, headings, and sections help retrieval.
• Ownership: every important source needs a content owner.
• Currency: old policies and duplicate copies reduce trust.
• Permissions: the agent should not expose content a user should not access.

Use Structured Business Data

Dataverse and other structured business sources help the agent answer questions that depend on records, status, relationships, and business context.

Structured data is different from document knowledge. A document may explain the leave policy. A Dataverse table may contain the employee's current leave balance, request status, or approval record. Together, they can produce a much better answer.

Use structured business data when the answer depends on:

• Current record status.
• Customer, employee, case, or asset details.
• Business relationships between records.
• Filtered data based on the signed-in user.
• Operational context, such as priority, region, owner, or due date.

Structured data plus business context helps the agent answer more accurately than a document-only approach.

Turn Documents Into Knowledge

Files are a practical way to bring existing knowledge into an agent. PDFs, Word documents, Excel files, and similar documents often contain policies, process guides, training notes, product sheets, and reference material.

Documents work best when they are written for retrieval:

• Use clear headings and subheadings.
• Keep one topic per section where possible.
• Remove outdated versions and duplicate copies.
• Avoid burying important answers inside screenshots only.
• Use plain language for common employee or customer questions.

The goal is to help users find answers without manually searching long documents.

One Agent, Multiple Sources

A production agent often needs more than one source. A support agent might use website documentation, SharePoint procedures, Dataverse ticket records, and uploaded troubleshooting guides. A policy agent might combine SharePoint policies, PDF handbooks, and structured HR data.

When combining sources, define which source should answer which kind of question. Otherwise, the agent may retrieve overlapping information from multiple places and produce inconsistent answers.

A simple source map helps:

• Policy explanation: SharePoint or document source.
• Current user status: Dataverse or business system.
• Public product help: website source.
• Internal procedure: SharePoint or controlled file source.
• Operational task: action or workflow, not knowledge alone.

Question, Retrieve, Answer

Knowledge-powered answering follows a retrieval-first pattern. The agent should find relevant information before generating the response.

1. Ask: the user asks a question in natural language.
2. Retrieve: the agent searches knowledge sources for relevant information.
3. Answer: the agent generates a grounded response using the retrieved information.

The rule is simple: retrieve first, generate second. This reduces vague answers and keeps the conversation tied to approved knowledge.

Knowing Is Not Doing

Knowledge sources answer questions. Actions perform tasks. These are related, but they are not the same.

Capability	Purpose	Example
Knowledge	Answers a question from trusted information	"What's our leave policy?"
Action	Runs a task or updates a system	"Submit my leave request."

A strong agent often uses both. It can explain the policy using knowledge, then submit a request using an action. Keep that boundary clear so users understand when the agent is informing them and when it is doing something on their behalf.

Access Still Matters

Connected does not mean accessible to everyone. Knowledge sources must respect identity, permissions, and content boundaries.

Before relying on a source, confirm:

• Who the user is.
• What the user is allowed to access.
• Which knowledge is relevant to that user's request.
• Whether the answer should include details, a summary, or a denied-access message.

This connects directly to Day 65. Knowledge source design and permission design must work together. A helpful answer is not helpful if it exposes information to the wrong audience.

Start With the Business Question

Do not start by connecting random sources. Start with the business question: What do users want to know?

Then work backward:

1. Business question: identify the real question users ask.
2. Trusted information: decide where the approved answer lives.
3. Right source: connect the most relevant knowledge source.

For example, a leave policy explanation may come from SharePoint, a leave balance may come from Dataverse, and public documentation may come from a website. Source quality directly affects answer quality.

Before You Publish

Good knowledge needs governance. Before publishing an agent, check the sources and the responses as carefully as you check the conversation flow.

• Trusted sources: use approved content, not random files.
• Current content: remove outdated or duplicate versions.
• Correct permissions: test access with real user roles.
• Real questions tested: use the language users actually type.
• Responses validated: confirm the answer is accurate, relevant, and safe.

Warning: Publishing an agent with messy knowledge can make bad content easier to find. Clean the source before scaling the conversation.

Key Takeaway

Knowledge sources are how a Copilot Studio agent finds trusted information and turns it into grounded responses. Websites, SharePoint, Dataverse, files, and enterprise data can all help, but only when they match the user's question and respect permissions.

The practical rule is: connect the right knowledge, validate the answers, and govern the sources over time. Better knowledge creates better agent experiences.

Next: Day 67: SharePoint as a Knowledge Source.