Building and customizing solutions using Microsoft 365 Copilot APIs and tools
Hi Shahana,
Great questions about Copilot Studio's knowledge source capabilities! Let me address both your concerns:
Table and Chart Reading Capabilities
- Current Limitations:
- Tables: Copilot agents can read text-based tables but struggle with complex formatting
- Charts/Images: Cannot read visual charts, graphs, or image-based content
- Complex layouts: May lose context in multi-column or nested table structures
- What Works:
- Simple HTML/markdown tables
- CSV-style data in text format
- Structured lists and bullet points
- What Doesn't Work Well:
- Excel embedded tables
- Image-based charts/graphs
- Complex SharePoint list views
- Visually formatted comparison charts
- Why This Happens:
- Semantic similarity: AI treats "GitHubCopilot" and "Copilot" as related terms
- Partial matching: Knowledge search finds articles containing both terms
- Context bleeding: Similar terminology across different articles creates confusion
- 1. Improve Knowledge Source Structure For better term differentiation: Article Titles: "Copilot Setup Guide" "GitHub Copilot Setup Guide - Developer Tool" "Using Copilot" "Using Microsoft Copilot Studio - Agent Builder"
- Add clear context in content:
- Use full product names consistently
- Add disambiguation in first paragraph
- Include product categories/contexts
- 2. Enhance Table Content for AI Reading Instead of complex tables, use: ## GitHub Copilot Registration Steps:
- 1. Step 1: Account Setup - Navigate to GitHub.com - Sign in to your account
- 2. Step 2: Subscription - Go to GitHub Copilot settings - Choose individual or business plan Rather than:
StepActionDetails1SetupNavigate...- 3. Use Structured Content Format
For comparison data:
## Product Comparison: GitHub Copilot vs Microsoft Copilot Studio
### GitHub Copilot (Developer Tool):
- Purpose: Code completion and generation
- Platform: IDE integration
- Cost: $10/month individual
### Microsoft Copilot Studio (Agent Builder):
- Purpose: Chatbot and agent creation
- Platform: Web-based builder
- Cost: Per conversation pricing
- 4. Implement Content Tagging Add metadata to your SharePoint articles:
- Tags: "github-copilot", "microsoft-copilot-studio"
- Categories: "developer-tools", "agent-building"
- Product-specific keywords in article metadata
- 5. Knowledge Source Organization Separate knowledge sources by product:
- Source 1: "GitHub Copilot Documentation"
- Source 2: "Microsoft Copilot Studio Guides"
- Source 3: "General AI Copilot Information"
- For Complex Tables/Charts:
- Convert to structured text before adding to knowledge base
- Use Power BI integration for data visualization queries
- Create text summaries of chart information
- Add table descriptions explaining what the data shows
- For Term Disambiguation:
- Create a glossary article defining different Copilot products
- Use prompt engineering in your agent instructions:"When users ask about 'Copilot', clarify which product they mean:- GitHub Copilot (code assistant)- Microsoft Copilot Studio (agent builder)- Microsoft 365 Copilot (productivity assistant)"
- Best Practices Content Structure:
- Use clear headings with product names
- Repeat key terms throughout content
- Add context clues in every section
- Knowledge Base Design:
- Separate similar products into different sources
- Use consistent naming conventions
- Include disambiguation content
- Agent Instructions:
- Train the agent to ask clarifying questions
- Set up confirmation flows for ambiguous requests
Thanks,
Karan Shewale.
*************************************************************************
If the response is helpful, please click "Accept Answer" and upvote it. You can share your feedback via Microsoft Teams Developer Feedback link. Click here to escalate.