Building and customizing solutions using Microsoft 365 Copilot APIs and tools
Issues with Copilot Studio Knowledge Base Accuracy from SharePoint & Missing Features (Enhanced Search, Deep Reasoning)
Hello Microsoft Community,
I am experiencing significant challenges with the accuracy of answers generated by my Copilot Studio bot when its knowledge base is connected to a SharePoint site. While the connection itself functions perfectly, the quality of responses is consistently low, often inaccurate or not directly addressing the query based on the SharePoint content.
My environment details:
- Licenses: I possess both Copilot Studio and Copilot for M365 licenses.
- Knowledge Base Connection: SharePoint site.
- Enabled Features: Tenant Graph Grounding, Semantic Search.
- Other Settings: Content moderation is set to 'low', authentication is configured for Microsoft authentication.
I have reviewed the Microsoft documentation on Copilot Studio knowledge bases, specifically https://learn.microsoft.com/en-us/microsoft-copilot-studio/knowledge-copilot-studio#tenant-graph-grounding. This document mentions key features that I believe could resolve my accuracy issues, but I cannot locate or enable them:
My questions are:
- "Enhanced Search Results" Setting: The documentation indicates an 'Enhanced search results' option that I need to enable for better grounding. However, I do not see this setting available in my Copilot Studio environment. Is this feature still in preview, limited by region/license, or has its location changed? What are the prerequisites to enable 'Enhanced search results' in my Copilot Studio bot, and where can I find this specific setting?
- "Deep Reasoning" Preview Availability: Similarly, the 'deep reasoning' preview mentioned in the documentation appears to be unavailable to me. Is this feature restricted to certain regions, tenants, or preview programs? How can I gain access to the 'deep reasoning' capability to improve response quality?
- Improving Overall Accuracy: Despite having enabled 'Tenant Graph Grounding' and 'Semantic Search,' and possessing both Copilot Studio and Copilot for M365 licenses (which, according to documentation, allows reading up to 22MB from SharePoint), the accuracy remains a critical concern. Are there any other configurations, best practices, or troubleshooting steps I should undertake to significantly enhance the accuracy and relevance of answers from my SharePoint knowledge base?
- Leveraging M365 License Benefits: Given that I have a Copilot for M365 license, which is stated to allow extensive SharePoint content reading, am I potentially missing any specific integration steps or settings within Copilot Studio that fully leverage this capacity for superior grounding and accurate responses?
Any guidance or steps to resolve these issues and improve my bot's accuracy would be greatly appreciated. Thank you