My Obsidian inbox had 40 fleeting notes (fragments from books, podcasts, conversations, and half-formed ideas). In the past I would set aside time to make them into proper Zettelkasten notes for my vault with 1000+ notes. I know that this is an important part of the Zettelkasten, but in the spirit of experimentation I wanted to see how AI could help with it. I tried delegating the process to AI, and it worked better than I expected.
Here’s the process I used.
- I started with a steering file. Before touching a single note, I wrote a steering file that defined exactly what a good note looks like: one idea per note, written as a direct timeless claim (not “this article says”), a headline that states the idea rather than the topic, five tags maximum, up to three cross-links to related notes, and a source field. The steering file also defined when to create a Map of Content (only when five or more notes cluster around a coherent theme).
- I let the AI read the inbox, then write to a separate folder. I’m not yet ready to test something new on my full Obsidian vault, so I limited the AI to a single folder with a test batch of 40 notes. I kept the folder read-only (so I could check work) and asked the AI to write notes to a separate folder. Claude read all 40 notes in one pass, identified which ones contained multiple ideas, and wrote each cleaned note from scratch in the Zettelkasten format. Three long research documents became seven separate notes. Four Maps of Content emerged naturally from the clusters that formed.
The result was 40 notes processed in one session, properly formatted, tagged, and linked. What I’m looking for ultimately is not the note rewriting, it’s the cross-linking and the ability to see patterns within notes. So I expect to continue testing beyond this first batch to explore that idea.
Leave a Reply