SharePoint 4 min read

Turn SharePoint into a Copilot Knowledge Base with Metadata

Copilot can reason over the custom metadata in your SharePoint libraries. Give a library the right columns and it becomes a knowledge base you can question in plain English, through Copilot in SharePoint or an agent of your own.

Key takeaways

  • Copilot reasons over custom metadata columns in SharePoint libraries, such as document status, owner and next review date.
  • With the right columns, questions like 'show me all expired documents' return accurate, filtered results.
  • A custom Copilot agent grounded in one library can combine columns, for example active documents grouped by subcontractor.
  • Without metadata, Copilot is guessing from file names and content.
  • Start with the libraries people query most and add the three or four columns they filter by.

To turn SharePoint into a knowledge base Copilot can use, give your document libraries metadata: columns for things like document type, status, owner and review date. Copilot reasons over those columns, so “which documents have expired?” comes back as an accurate, filtered list instead of a vague summary. Without them, Copilot is guessing from file names and content.

For years metadata was the thing people skipped because folders felt easier. With Copilot reading your libraries, the columns you add now decide which questions it can answer.

Why metadata decides what Copilot can answer

A Document Status column with values like Active and Expired, and a Next Review Date on every document, gives Copilot facts to filter on instead of text to interpret.

My demo used a project site with three libraries: compliance documents, permits and approvals, and client communications. The compliance library carried columns for document type, status, owner, next review date and subcontractor, and every result below comes from those columns.

Asking a library questions with Copilot in SharePoint

When I recorded this, the button in the corner of the library was the Knowledge Agent preview. That preview has since become Copilot in SharePoint, the floating Copilot button licensed users see in their sites.

“Get me all the expired documents” returned four documents. Filtering the library’s Document Status column by Expired showed the same four.

“Documents with a next review date this month” also returned four, and filtering Next Review Date for the month showed the same four. Copilot worked out what “this month” meant on its own. Nobody had to open the filter pane or know which column held the answer.

Building a Copilot agent on a single library

The second approach is an agent grounded in one library. My Compliance Copilot agent was built with the lighter agent builder rather than full Copilot Studio, and pointed at the compliance documents library only.

From Microsoft 365 Copilot, without opening SharePoint, it returned the same four expired documents in a table and the same four review-date matches.

The harder test combined two columns: summarise the document types with a status of Active, grouped by subcontractor. It came back with three active documents across two subcontractors, two for Greencore Civil and one for Precision, with notes against each. Filtering the library by status confirmed all three.

Start with one library

  1. Pick the library people already ask questions about. Compliance documents, contracts, policies and project records. If your team keeps emailing each other “do we have a current insurance certificate for…?”, that’s the library.
  2. Add three or four columns to start. Usually a type, a status, an owner and one date that drives action, like review or expiry. Every extra column is one more thing people have to fill in, and a column that’s half filled in means Copilot misses documents.
  3. Use choice columns for anything with set values. Status should be Active, Expired or Under review, not whatever someone typed that day.
  4. Don’t retag the whole tenant. Pick one library, get it right, and test it with the real questions your team asks. Autofill columns and default values will clear most of the backlog.
  5. Check permissions before you go live. Copilot only returns what the person asking can already open. If an agent surfaces a document someone shouldn’t see, the problem was the library’s permissions, and it was there before Copilot.

For the wider foundations, why your SharePoint needs an information architecture before AI covers site structure and oversharing. I also teach the full process, from building the library to putting an agent on top of it, in Copilot Ready Libraries.

Frequently asked questions

Can Copilot filter SharePoint documents by date? Yes. In my demo, “documents with a next review date this month” returned the four documents whose Next Review Date fell in the current month, matching the library’s own filter.

What happened to the SharePoint Knowledge Agent? It was the preview that let you ask a SharePoint library questions in plain English. It has since become Copilot in SharePoint, which appears as a floating Copilot button for users with a Microsoft 365 Copilot licence.

Should I use choice columns or text columns for metadata? Use choice columns wherever the values are fixed, like status or document type. Free text gives you Expired, expired and EXP, and your filters won’t group them.

Should I use a Copilot agent or Copilot in SharePoint? Copilot in SharePoint is quickest when you’re already in the site. An agent grounded in one library suits a team that asks the same kind of question often, because they can ask from Microsoft 365 Copilot without opening SharePoint.

Get these in your inbox. One email a week.

Practical thinking on organising your information and getting real work out of AI. Subscribers get the free SharePoint Skill course to start.