Use SharePoint as a knowledge source in Copilot Studio

In Copilot & AI · 9 July 2026

Copilot Studio Power Platform How-to SharePoint

Copilot Studio can use SharePoint as a knowledge source, which lets your agent answer questions from the document libraries and lists your organisation already keeps there rather than from its general training. That makes it one of the quickest ways to ground an agent in content your teams already maintain. The setup is short, but the decisions that come first are what matter: whether to use documents or lists, which retrieval method to use, and how to handle authentication.

How to use SharePoint content for your agent

Copilot Studio supports two types of SharePoint content as knowledge sources: document libraries and lists.

Document libraries are the most common starting point. Agents can access Word, PDF, and PowerPoint files. The file size limit depends on your search method: up to 7 MB per file with standard SharePoint search, or up to 200 MB with Tenant Graph Grounding with Semantic Search (more on that below). Excel and CSV files are not supported.

SharePoint Lists are also available as a knowledge source (production-ready preview as of July 2026). Text, choice, number, date, and lookup columns are indexed. Calculated columns are not, so the agent cannot use them.

The quality of what the agent returns depends on the quality of the content itself but also on how you manage the data. A library of scanned PDFs with a poor text layer, or a list with vague column names, will produce poor results regardless of which search method you choose.

Standard SharePoint search vs Tenant Graph Grounding with Semantic Search

Before you add any knowledge source, decide which retrieval method your agent will use. The difference affects response quality, file size limits, licence requirements, and cost.

The option to enable semantic search is under Settings > Generative AI: look for Tenant graph grounding with semantic search. It requires generative orchestration to be enabled on the agent first.

Semantic search works better for internal agents used by personnel with a Microsoft 365 Copilot licence. Standard search is fine for external-facing agents, tightly scoped document sets such as HR policies, or where the semantic search cost is not justified.

Add a SharePoint document library as a knowledge source

With your search method decided and authentication configured (see the permissions section below), go to the agent’s Overview page and select Add knowledge, or go to Knowledge and select + Add.

Select SharePoint from the list. Enter the site URL in this format:

https://yourorganisation.sharepoint.com/sites/your-site-name

Fill in the Name and Description fields. Generative orchestration uses both to decide which knowledge source to query when a message comes in. A sentence that describes the content accurately works best: something like “IT support policies, approved software lists, hardware request procedures.” A vague label like “IT SharePoint” causes the agent to miss relevant queries. In practice this is one of the most common mistakes I see: users assume the language model will work out what each source contains on its own, so they leave the description vague and point the agent at hundreds of files.

Select Add to agent.

SharePoint Lists as a knowledge source

From the Knowledge page, select + Add > SharePoint, then choose Browse items to see My Lists and Recent Lists, or paste the list URL directly:

https://yourorganisation.sharepoint.com/sites/your-site-name/Lists/your-list-name

The hard limit is 15 lists per agent; Microsoft recommends staying at 10 or fewer for best results. Lists with more than 35,000 rows reduce quality and increase latency, and total rows across all lists should not exceed 120,000.

When to use a list vs a document: lists work well for structured, frequently updated reference data: approved suppliers, software catalogues, project registers, equipment inventories. Documents work better for policies, procedures, and anything narrative. In practice you will often use both in the same agent.

Test the agent before you publish

Open the Test panel and ask a question the SharePoint content should answer. A working knowledge source produces a response with a citation card linking to the source document or list item.

If the agent says it does not know:

1. Check authentication first. This is the most common cause of silent failures (see below).
2. Check the knowledge source description. Too vague, and the orchestrator will not select it for the query.
3. Check the content itself. Confirm the information sits in a supported file type, that scanned documents have a real text layer, and that the relevant list column is not a calculated field.

Test with at least three questions that should draw from different documents or list rows. Citation cards on all three confirm the knowledge source is working correctly.

Permissions and security for your agent

Authenticate with Microsoft is required for any agent that accesses private SharePoint content. Set it at Settings > Security > Authentication > Authenticate with Microsoft before you add your knowledge source. Without it, the agent cannot read the SharePoint site and returns nothing, with no error message to indicate why. If you are new to how identity works across Microsoft’s cloud, Microsoft Entra ID for Dynamics 365 covers the fundamentals.

What happens if a user does not have read permission to a SharePoint document or list item: the agent respects SharePoint’s existing security trimming, so it never shows content a user cannot open directly. If someone asks about content they do not have permission to read, the agent says it does not know, or draws from other sources they can access. For agents deployed across multiple departments, I recommend explaining this in the welcome message so users understand why they might get different answers.

For more complex scenarios, including agents that need to access SharePoint on behalf of different users or via a service account, refer to the advanced authentication scenarios in the Microsoft documentation.

Where to go next

A knowledge source turns your agent into something users can query. The next step is making it act on its own: building an autonomous agent with event triggers lets you respond to changes in SharePoint or Dataverse without waiting for a user prompt.

If you are preparing for a certification, the AB-620 covers agent design, knowledge sources, and deployment, and the PL-200 covers the broader Power Platform skills you will use alongside Copilot Studio.