How do you export a knowledge base for AI licensing?
Late, and once. Inventory the spaces first, agree scope with a buyer, redact, then export only the agreed spaces in a format that keeps links, titles, owners and dates. Nothing needs to leave your systems before a buyer has scoped it.
Written November 7, 2026. Menu names change; check each tool's help pages.
When should you export?
After three things are settled: which spaces or folders are in scope, who removes personal details and when, and what format the buyer can read. Exporting earlier creates a full copy of everything, names included, sitting on someone's laptop.
Before that, count. A one-line inventory per space (page count, oldest edit, owning team, whether it names people) is what a buyer needs to scope the deal.
How do Notion, Confluence, Google Drive and SharePoint export?
| Tool | Whole-workspace route | Formats that keep structure | Watch for |
|---|---|---|---|
| Notion | Workspace settings, export all workspace content (owner rights) | Markdown and CSV, or HTML; databases come out as CSV | Database properties holding people and emails |
| Confluence | Space settings, export space (space admin rights) | HTML or XML; PDF loses links | Inline comments, attachments, personal spaces |
| Google Drive | Admin console Data Export for the organization; Google Takeout for one user | Docs as Word or HTML, Sheets as Excel or CSV | Comment threads and suggestion history in Docs |
| SharePoint and OneDrive | No single export; sync or download libraries, or an admin export through Microsoft Purview | Original Office files, with a list export of library metadata | Version history, "modified by" columns, linked Teams files |
What should an export keep, and what should it drop?
| Keep | Drop |
|---|---|
| Page titles and folder tree | Personal and private spaces |
| Links between pages | Comments and suggestion threads |
| Created and edited dates | "Created by" and "edited by" names |
| Labels and page templates used | Attachments not yet checked |
| Version history, if the buyer wants it and it is cleaned | Client and third-party files |
Scale AI lists "documents and files with metadata" among what it takes, which is why dates, titles and structure are worth keeping.
Sources, with read dates
- scale.com/data-partnership (opened 11/7/2026)
- micro1.ai/data (opened 11/7/2026)
- simpleclosure.com/product/assethub (opened 11/7/2026)
Questions on file
- Which format do buyers prefer?
- None of the buyer pages we read names a file format. Formats that keep structure (HTML or XML with links, Markdown with folders, a CSV index of page titles, owners and dates) carry more context than flat PDFs. Ask the buyer before you export.
- Should we include attachments?
- Only after checking them. Attachments are where client files, scanned forms and screenshots hide. Leave them out of the first export unless the buyer asks.
- Do we need an admin to export?
- For a whole workspace, usually yes. Notion workspace exports and Google Workspace data exports are admin or owner tasks, and Confluence space exports need space admin rights.
- Can the buyer connect to our tools instead?
- Sometimes. Scale AI, for one, does its anonymizing on your systems ahead of any extraction. If a buyer is to have direct access, the agreement should define it, and define the day it stops.
Get a quote
Tell us where the documents live and roughly how many there are. Keep them where they are; a description is enough for us to check which buyers fit.
Note: we make introductions, and the buyer at the other end may pay us for them.
Keep documents, exports, share links and samples to yourself. We work from your description alone.
Filed. Next steps:
- Your answers get compared with every buyer's posted size, region and content rules.
- An email follows, naming each buyer that fits and why, or telling you no buyer fits so far.
- The company's name reaches a buyer only after your yes to that particular buyer.
- Until then, leave the files exactly where they are. Copies are not needed.