THE ZETTELKASTEN MANIFESTO | What is a Zettelkasten?

Zettelkasten systems work best when notes behave like flexible “containers” linked by lightweight identifiers—not when every note is forced into a rigid, one-size-fits-all template. The core problem behind most confusing advice is structural: many tutorials push strict note formats (IDs, summaries, outlines, bullet lists, and specific ordering of content), but real research notes often include wildly different media—quotes, images, audio, video, references, and inconsistent metadata. Niklas Luhmann’s original practice is used as the anchor example, then translated into a more general data model that can handle modern, mixed content.

Luhmann’s method is described as a concrete system with three essential parts on each paper slip: a sequential number (an ID), the note’s content (thoughts, quotes, references), and a set of numbers pointing to other notes. Those internal references let him retrieve related ideas by following links. The system functioned like a manually maintained database: each note was a record, and the references acted like pointers to other records. The transcript emphasizes that Luhmann didn’t have automated citation tools or software databases, yet his approach still mirrors how contemporary relational databases store and connect information.

But relational thinking breaks down once notes stop being purely text. If every note must fit into table columns—one column for “duration,” another for “references,” another for “image metadata,” another for “sound description”—then the system becomes brittle. A single “additional information” column also fails because automation and searching become ambiguous: should “4 minutes 30 seconds” be treated as a duration or just a string? Should a reference be parsed differently depending on whether it appears with a header or not? The transcript argues that these structural constraints quickly create trouble when users want to search by media type (e.g., only videos) or when metadata is uneven across notes.

The proposed alternative is a graph database mindset. Instead of forcing everything into one table with many columns, each note becomes a node-like “box” that can hold whatever content it contains—text, an image, a video, audio, or combinations—while only minimal metadata (like tags and IDs) is attached. Links between boxes are “weak” at first: searching by text or tags can return multiple related boxes, and then additional “hard” links (explicit IDs referencing other boxes) can be created when relationships become clear. This supports flexible retrieval: an image box and a video box can both be returned for the same search term, and the system can later connect them more directly.

The takeaway is practical: don’t copy a rigid Zettelkasten template. Build a personal system that treats notes as an open-ended set of linked containers, where the only consistently required elements are identifiers, content, and relationships. The transcript frames this as the path to a Zettelkasten that remains useful as the variety and messiness of real research data inevitably grows.