Sources: Your Window Into AI Decision-Making
Sources are your direct insight into which documents and knowledge your AI used to craft each response. Found beneath every AI message in your conversations (look for the book icon), Sources show you exactly what information informed each answer - transforming your AI from a "black box" into a transparent, accountable assistant.
Why Sources Matter
Every AI response now comes with receipts. When your AI tells a customer about your return policy, Sources shows you it referenced "Customer Service Guidelines". When it quotes pricing, you'll see "Q4 Pricing Sheet - Premium Tier." No more wondering where information came from.
Sources act as your quality control system. Before any response goes out, you can quickly verify:
- Did the AI use the right documents?
- Are the sources current and accurate?
- Is anything missing from the knowledge base?
🛠️ Troubleshooting Made Simple
When a response doesn't match your expectations, Sources tell you exactly why:
Problem: AI gave outdated pricing information
Sources Investigation: Shows AI referenced "2023_pricing.pdf"
Solution: Remove old pricing document, update with current rates
Problem: AI couldn't answer a product question
Sources Investigation: No sources listed = no relevant documents found
Solution: Add product documentation to knowledge base
Problem: AI gave incomplete warranty information
Sources Investigation: Only referenced "basic_warranty.pdf"
Solution: Ensure AI can access "extended_warranty.pdf" and "premium_coverage.pdf"
Real-World Troubleshooting Examples
Case 1: Inconsistent Policy Information
-
Customer Complaint: "Your AI told me 30-day returns, but your website says 60 days!"
-
Investigation: Check Sources for that conversation
-
Finding: AI referenced "old_policy_2023.pdf"
-
Fix: Remove outdated document, upload current policy
-
Result: Consistent 60-day return information across all channels
Case 2: Missing Product Features
-
Issue: AI not mentioning new security features to enterprise customers
-
Investigation: Sources show AI only used "product_overview_mar_2025.pdf"
-
Finding: Enterprise feature documentation exists but not linked to campaign
-
Fix: Add enterprise documentation to campaign knowledge sources
-
Result: AI now highlights security features for enterprise inquiries
Case 3: Pricing Confusion
-
Problem: AI giving different prices to similar customers
-
Investigation: Sources reveal AI alternating between "standard_pricing.xlsx" and "promotional_pricing.xlsx"
-
Finding: Both documents exist with conflicting information
-
Fix: Clarify in Notes when to use promotional pricing, update documentation
-
Result: Consistent pricing logic across all customer interactions
How to Use Sources Effectively
📋 Regular Review Process
- Weekly Source Audits: Review frequently used sources in your conversations
- Gap Analysis: Note when no sources appear for common questions
- Accuracy Checks: Verify that referenced sources contain current information
- Coverage Assessment: Ensure all business areas have proper documentation
🔧 Proactive Troubleshooting
- No Sources Listed: Add relevant documents to knowledge base
- Wrong Sources Used: Review document titles and content for clarity
- Outdated Sources: Replace old documents with current versions
- Incomplete Sources: Check if related documents are properly connected
📈 Knowledge Base Optimization
Sources data reveals which documents are most valuable:
- Frequently Referenced: Keep updated and easily accessible
- Never Referenced: Review relevance or improve discoverability
- Missing for Common Questions: Create new documentation
Building Trust Through Transparency
Sources transform customer support from "trust us" to "here's proof." When customers question information, you can point to specific sources. When internal teams need to verify AI responses, Sources provide immediate validation. This transparency builds confidence in your AI system and helps teams trust automated responses.
- Check Out Sources View: Click the book icon below any AI response in your conversations
- Start Monitoring: Review sources for responses that seem off or incomplete
- Document Gaps: Note when important questions have no sources listed
- Iterate and Improve: Use insights to enhance your knowledge base
Sources aren't just about accountability - they're your roadmap to building a smarter, more reliable AI assistant that your team and customers can trust.
Douglas Ho
Comments