Copilot Studio Agent: SharePoint Excel knowledge source (using Sharepoint connector) no longer returns results since this week

Question

Hello,

I'm experiencing an issue with a Copilot Studio Agent that was working correctly until Wednesday morning.

Service

• Microsoft Copilot Studio (Standard harnass)
• SharePoint Excel file used as a knowledge source (connected via the Sharepoint connector)

Scenario

We have an agent called Catalog Checker that searches a catalog stored in an Excel file on SharePoint. Since Wednesday morning, the agent appears to no longer read the Excel file correctly. For every product that is queried, the agent responds that the product cannot be found in the catalog.

Troubleshooting Performed

• Verified that no changes were made to the agent instructions.
• Verified that no changes were made to the knowledge source configuration.
• Tested with different models for the agent.
• The behavior remains the same regardless of the selected model.
• The solution was still working correctly on Monday.
• Finally removed and added the knowledge source again, without any results.
  **
  Question
• Could there have been a recent change in the model configuration, knowledge retrieval behavior, or SharePoint Excel knowledge source processing for Standard Copilot Studio agents that could explain this behavior?
• Has anyone experienced similar issues with SharePoint-hosted Excel knowledge sources suddenly no longer being queried correctly?

Thank you for any guidance.

Answer 1

This is unlikely to be caused by the selected model because SharePoint retrieval happens before response generation. Please confirm how the Excel file was added.

If it uses the standard SharePoint site/URL connector, reliable Excel cell retrieval isn’t documented; Microsoft recommends Code interpreter for Excel. Alternatively, add the file through Upload files > SharePoint, wait until its status returns to Ready, and allow the 4–6-hour synchronization period.

Also verify that the querying user can find the product using SharePoint search, has direct Read access, and that the file isn’t encrypted, password-protected, or sensitivity-protected.

If a small supported XLSX file still fails after enabling Code interpreter and Work IQ, publishing, and testing with multiple authorized users, this should be reproduced and raised as a potential regression.