From OneNote to Copilot: Building an AI-Ready Knowledge Base in One Morning

A practical case study on transforming unstructured OneNote content into a structured, AI-consumable knowledge base—without custom development.

🧠 The Problem: Knowledge Exists… But Isn’t Usable by AI

Many organizations rely heavily on OneNote as a living knowledge base. It works extremely well for human consumption—but breaks down when you try to operationalize that data for AI systems like Copilot.

Hundreds of pages across multiple sections
Critical procedures and troubleshooting content
No structured indexing beyond native search
No usable integration point for Copilot or AI agents

The real challenge becomes: How do you make OneNote content structured and queryable—without building a full ingestion pipeline?

⚡ The Approach: Keep It Simple, Move Fast

Instead of introducing heavy architecture or custom code, this solution uses Power Automate as a lightweight ETL pipeline.

Key Idea: Convert OneNote pages into structured HTML and store them in SharePoint—where Copilot can actually consume them.

🔧 What the Flow Does

Enumerates notebook sections
Loops through every page
Extracts full HTML content
Writes files to SharePoint in a structured hierarchy

🧩 Architecture Overview

OneNote (Source)
   ↓
Power Automate Flow
   ↓
OneDrive (Staging)
   ↓
SharePoint (Structured KB)
   ↓
Copilot / AI Consumption

🔍 Implementation Breakdown

1. Manual Trigger

The flow is intentionally manual—this keeps execution controlled and avoids unnecessary cycles.

2. Enumerate Sections

Uses GetSectionsInNotebook to dynamically pull all sections from the source notebook.

3. Iterate Sections

A loop ensures the solution scales with notebook growth without modification.

4. Enumerate Pages

Each section is expanded using GetPagesInSection to ensure full coverage.

5. Extract Content

GetPageContent retrieves full HTML—preserving formatting and structure.

6. Stage Files

Pages are written to OneDrive as temporary HTML files.

7. Publish to SharePoint

Files are then stored in SharePoint using a structured folder hierarchy that mirrors the notebook layout.

✅ Why This Works

No custom development required
Built in a single morning
Uses only native Microsoft 365 connectors
Produces AI-indexable content immediately

Outcome: The entire knowledge base becomes searchable and usable by Copilot with zero replatforming.

📚 Key Lessons Learned

Unstructured data limits AI effectiveness
Power Automate is a viable ETL-lite platform
Speed beats perfection for initial delivery

⚠️ Known Limitations

Reprocesses all content every run
No change tracking or deduplication
Double write (OneDrive + SharePoint)
No built-in error handling

These trade-offs are acceptable for:

Initial ingestion scenarios
Proof-of-concept deployments
Rapid enablement projects

🚀 Next Evolution Opportunities

Incremental sync (based on page timestamps)
Duplicate detection
Direct SharePoint writes (remove staging)
Error handling and logging
Metadata tagging for better Copilot responses

💼 Business Impact

✅ Enterprise knowledge made AI-accessible
✅ Faster internal troubleshooting and support
✅ Enables Copilot-driven workflows
✅ Avoids costly custom development

🎯 Final Thought

This is a clean example of pragmatic engineering:

Simple architecture. Fast execution. Immediate value.

You don’t need a full data platform to unlock AI value—sometimes you just need the right connector and a few well-placed loops.

Blog