The Zettelkasten method: how it works, the evidence and when not to bother

Luhmann and the original box

Niklas Luhmann, the German sociologist, kept a card index from 1951 until shortly before his death in 1998. According to Johannes Schmidt, who leads the edition project at the Niklas Luhmann-Archiv in Bielefeld, it consists of approximately 90,000 handwritten A6 cards in two collections: the first, created between 1951 and 1962, holds about 23,000 cards in 108 subject sections with a keyword index of roughly 1,250 entries; the second, kept from 1963 to 1997, holds about 67,000 cards in eleven top-level sections with around 100 subsections and a keyword index of 3,200 entries.^{[1] [2]} A Luhmann described the system in a short 1981 essay, "Kommunikation mit Zettelkästen", and called it a "second memory" and a communication partner that could surprise its author.^{[2] [3]}

Schmidt's descriptions are the most careful we have of how it worked.^{[1] [2]} Four features matter, and they are not quite what the modern method says.

1. Fixed placement, branching numbers. A card, once numbered, never moved. When a thought on an existing card led somewhere, Luhmann wrote the new card and inserted it directly behind the one it continued, giving it a branching number; the numbers alternate digits and letters and run to thirteen characters, as in 21/3a1p5c4fB1a.^[2] A The result is a sequence that follows the connectivity of arguments, not a taxonomy, and Schmidt notes that "there are no privileged positions in this web of notes: there is no top and no bottom".
2. References between cards. Cards carry the numbers of other cards elsewhere in the collection, often in a completely different subject area; Luhmann called it a "web-like system" and added references to old cards when new ones made them relevant.^[2] A These are what the digital world calls links.
3. A deliberately sparse keyword index. The index makes no claim to list every card on a term. Luhmann "typically listed only one to four places where the term could be found", on the idea that the internal references would lead to the rest.^[2] A
4. Unplanned combination as the point. Schmidt titles his article "the fabrication of serendipity": because references can lead to a distant region of the file, following them turns up neighbours you did not look for.^[2]

Luhmann's list of publications ran to more than 500 titles at his death.^[2] Whether the box caused that output or recorded the thinking of a very productive person is not something any study can settle; the archive shows a working tool of one scholar, not a validated method. That is where the popular literature begins.

Note types and linking rules: the modern method

The version most people learn comes from Sönke Ahrens's 2017 book How to Take Smart Notes, which translated Luhmann's practice into a workflow for students and writers.^[4] Its core rules:

• Fleeting notes capture a thought in passing and are discarded once processed.
• Literature notes record, in your own words and briefly, what a source says and where.
• Permanent notes (often called main or Zettel notes) each hold one idea, written as a full thought someone else could understand, and are placed in the box by linking them to existing notes.
• Atomicity: one idea per note, so that it can be linked from several directions.
• Write for the connection: when filing a note, ask which existing note it continues, contradicts or qualifies, and write that reason into the link.
• Index sparingly: a few entry points into the box, not a tag on everything.

The tripartite note taxonomy is Ahrens's teaching device; Luhmann's cards do not come labelled that way, though Schmidt confirms that Luhmann kept reading notes apart and integrated them into the collection later.^[2] The rules about writing for the connection and indexing sparingly are faithful to the archive. The claim that the method makes writing "effortless" is the book's, not the archive's.

Demand for the method is steady rather than growing: "zettelkasten" draws about 6,600 US searches a month (keyword difficulty 28), down 18% year on year with a three-year compound growth rate of 5%; the UK adds 1,300, Germany 2,900 (up 26% in a year, difficulty 10), France 880, Spain 590 and the Czech Republic 210 (DataForSEO, retrieved 2026-11-09). "obsidian note taking app" at 2,900 a month shows where the practice has moved.

What memory research supports

The method makes several implicit claims about learning. Some are well supported, some are supported for a different reason than the method gives, and some have no evidence either way.

Claim built into the method	What the research says	Grade
Writing a note in your own words beats copying or highlighting	Highlighting and rereading are rated low-utility techniques; elaborative interrogation and self-explanation, which is what rewriting in your own words involves, rate moderate, with the strongest evidence for practice testing and spaced practice[5]	A
Having to retrieve what you know in order to connect a new note strengthens memory	In a direct comparison, a single retrieval-practice session produced more learning a week later than elaborative studying with concept mapping, which is the activity closest to building a note network[6]. Linking a note forces retrieval of the notes it connects to; the gain comes from the retrieval, not from the map	A
Returning to old notes over time deepens understanding	Spaced practice beats massed practice across 254 studies, and the optimal gap grows with how long you need to remember[7]. A box you revisit while writing is a spacing schedule by accident	A
Externalising your thinking frees your mind	Cognitive offloading is adaptive and frees capacity, but it lowers unaided recall of what was offloaded and people are not well calibrated about when to do it[8]. A box you can search is a reasonable trade; a box you believe you remember is not	A
The network of links produces new ideas	No controlled evidence either way. Schmidt documents that it worked for Luhmann[2]; whether link density in a personal note store predicts creative output has not been tested	C
Writing notes this way makes a book write itself	A claim of the popular literature,[4] untestable as stated and contradicted by the archive's evidence of decades of effort	C

So the honest summary is this. The parts of the method that involve doing something with a text, rewriting it, retrieving what you already hold to connect it, and coming back to it, are supported by some of the most robust findings in learning research.^{[5] [6] [7]} The parts that involve the structure of the store, link density, the graph, the sparse index, are supported by one scholar's working life and no experiment.

Where it is folklore

• "Luhmann's productivity came from the box." Correlation in a sample of one. The archive shows the tool; it cannot show the counterfactual.
• "The graph view reveals hidden connections." Luhmann had no graph and found his connections by following sequences and references. A graph of a few thousand notes is a hairball; nobody has shown that looking at one produces an idea.
• "Atomic notes are the key." Luhmann's numbering was built for continuing a thought across many cards, not for isolating one idea per card.^[2] Atomicity is a useful discipline for linking; it is not a finding.
• "You will forget everything you don't put in the box." The Google-effect experiment that popularised the fear of digital forgetting failed to replicate in a large preregistered project;^{[9] [10]} offloading has costs, but they are specific to the offloaded item and manageable.^[8]
• "Folders are wrong, only links are right." Luhmann's second collection had eleven top-level sections and about a hundred subsections, and the branching numbers are themselves a hierarchy.^[2] He used both; so can you.
• "AI makes the Zettelkasten obsolete" or "AI completes it." Both are projections. The one piece of evidence that bears on it is indirect: knowledge workers who trusted generative AI more reported putting less critical-thinking effort into its output.^[11] C If the point of the box is to make you think, a tool that writes the notes for you removes the mechanism the research supports.

A minimal workflow in Obsidian

Obsidian stores notes as plain Markdown files on your own disk, supports links between notes natively and is free for personal use. We have no commercial relationship with it. The setup below is deliberately small; for the full tool comparison see Obsidian vs Notion.

Folders (three, no more):

• inbox for fleeting and literature notes, cleared weekly.
• notes for permanent notes, flat, no subfolders.
• index for a handful of entry-point notes, one per theme you actually write about.

Note template (set as the default template in the core Templates plugin):

# Claim in one sentence
Body: the idea in your own words, 3–8 sentences.
Because: [[note-this-continues]] — why it connects.
Against: [[note-this-qualifies]] — if any.
Source: author, year, page.

Daily, ten minutes. Move one inbox item to notes, rewriting it from memory before you look at the source again (that is the retrieval step^[6]). Then open one existing note it relates to and add the link with the reason. Do not add a note you cannot link.

Weekly, twenty minutes. Delete the rest of the inbox without guilt. Open one index note and read the chain of linked notes beneath it; where two notes say the same thing, merge them; where they disagree, write the disagreement as a new note.

Plugins: none required. The core Backlinks pane shows what points at a note. Skip the graph view until you have a few hundred notes, and treat it as decoration then.

A minimal workflow in Notion

Notion can hold the same system as a single database, which suits people who already live in it and want their notes beside their projects. The trade-off is that notes are rows in a hosted service rather than files you own; the second brain guide covers that trade and the export routine.

• One database, Notes, with properties: Type (select: fleeting, literature, permanent), Connects to (relation to the same database), Why (text, the reason for the connection), Source (text), Reviewed (date).
• A Permanent view filtered to Type = permanent, sorted by last edited, and a Review view filtered to Reviewed more than 30 days ago, which is your spacing schedule.^[7]
• The same daily and weekly routine as above. Notion's AI features can summarise a page; for the reasons in the folklore section, write the note yourself and use the summary only to check you have not missed a point.

Common mistakes

Mistake	Why it fails	Fix
Capturing everything	The inbox becomes an archive; nothing is rewritten, so the retrieval step never happens[6]	Capture only what you expect to write about; delete the inbox weekly
Tagging instead of linking	Tags are a taxonomy; they do not carry the reason for a connection, which is the part that makes you think	One link with a "because" beats five tags
Building the system before the notes	Templates, plugins and folder debates substitute for reading	Three folders, one template, start
Linking everything to everything	Dense links make every path equally uninformative; Luhmann's keyword index listed one to four places per term on purpose[2]	Link to the note a thought continues or contradicts, rarely more than two
Treating the box as memory	Offloading reduces unaided recall of what you offload[8]	Reread the chain under an index note when you start a piece of writing
Letting AI write the notes	Removes the rewriting and retrieval the research supports[5] [6]; associated with less self-reported critical effort[11]	Write first, then ask the model what you missed

When a folder is enough

The method is a writing tool. It pays off when you produce long-form arguments from many sources over years, which describes researchers, non-fiction writers, some lawyers and analysts, and few other people. If your notes are mostly meeting records, how-to snippets, project context and reference material, you are not building an argument and the linking discipline is overhead.

For that majority, a flat folder of dated Markdown files with good search, or a notes app with a decent index, meets the real need: finding the thing again. Our personal knowledge management tools guide compares the options on exactly that criterion, and the second brain method, built around projects rather than ideas, fits project-shaped work better than a Zettelkasten does.

Two questions decide it. Do you reread your notes to write something, at least monthly? Do you ever find a note that changes your mind about another? Two yeses and the method will earn its time. Otherwise, a folder and a search box, and the hour saved each week goes to reading.

What to do next

If you answered yes twice, build the three-folder Obsidian vault above in ten minutes; the template and the two routines are the whole system, and we send a packaged starter vault with the index notes and the review checklist to briefing subscribers. Then use the personal AI operating system builder to decide where the vault sits in the rest of your setup: what feeds it, what it feeds, and which AI layer is allowed to search it without writing into it.

Frequently asked questions

What is the Zettelkasten method?

A note-taking system modelled on Niklas Luhmann's card index: each note holds one idea written in your own words, is given a fixed place and is linked to the existing notes it continues or contradicts, with a sparse index of entry points. Modern versions (Ahrens 2017) add fleeting and literature notes as stages before a permanent note.

How many notes did Luhmann's Zettelkasten have?

Approximately 90,000 handwritten A6 cards in two collections, according to Johannes Schmidt of the Niklas Luhmann-Archiv: about 23,000 in the first collection (1951–1962) and about 67,000 in the second (1963–1997), plus keyword indexes of roughly 1,250 and 3,200 entries. Bielefeld University has digitised the cards and publishes them online.

Does the Zettelkasten method actually work?

The parts that involve doing something with a text are well supported: rewriting in your own words (elaboration), retrieving what you know to connect a new note (retrieval practice) and returning to notes over time (spacing). The parts about structure, such as link density, the graph view or atomic notes, rest on one scholar's practice and have not been tested. Whether it works for you depends on whether you write long-form arguments from many sources.

Is Obsidian or Notion better for a Zettelkasten?

Obsidian fits the method more naturally: local Markdown files, native links and backlinks, free for personal use. Notion works as a single self-related database if you already run your projects there and accept hosted rows instead of files. Our Obsidian vs Notion comparison covers data ownership, linking, AI features and cost.

Zettelkasten or second brain: which should I use?

A Zettelkasten is organised around ideas and serves writing; a second brain (PARA) is organised around projects and serves getting things done. If your notes are meeting records, snippets and project context, the second brain structure or a plain folder fits better. If you reread notes monthly to write and sometimes find a note that changes your mind about another, the Zettelkasten earns its time.

Should I let AI write my Zettelkasten notes?

Not the notes themselves. The learning benefit comes from rewriting and retrieving, which an AI summary removes; the only study of the question so far, a survey of 319 knowledge workers, found that greater trust in generative AI was associated with less self-reported critical-thinking effort. Write first, then use the model to check what you missed or to search the vault.

Sources and evidence

1. Niklas Luhmann's Card Index: Thinking Tool, Communication Partner, Publication Machine (in: Forgetting Machines: Knowledge Management Evolution in Early Modern Europe) — Schmidt, Johannes F. K. (2016), book A doi.org/10.1163/9789004325258_014 (retrieved 2026-11-09)
2. Niklas Luhmann's Card Index: The Fabrication of Serendipity (Sociologica 12(1), pp. 53–60; full text read) — Schmidt, Johannes F. K. (2018), observational A doi.org/10.6092/issn.1971-8853/8350 (retrieved 2026-11-09)
3. Kommunikation mit Zettelkästen. Ein Erfahrungsbericht (in: Öffentliche Meinung und sozialer Wandel; the digitised cards are published by the Niklas Luhmann-Archiv, Bielefeld University) — Luhmann, Niklas (1981), book B niklas-luhmann-archiv.de/ (retrieved 2026-11-09)
4. How to Take Smart Notes: One Simple Technique to Boost Writing, Learning and Thinking — Ahrens, Sönke (2017), book C (retrieved 2026-11-09)
5. Improving Students' Learning With Effective Learning Techniques: Promising Directions From Cognitive and Educational Psychology — Dunlosky, J.; Rawson, K. A.; Marsh, E. J.; Nathan, M. J.; Willingham, D. T. (2013), systematic review A doi.org/10.1177/1529100612453266 (retrieved 2026-11-09)
6. Retrieval Practice Produces More Learning than Elaborative Studying with Concept Mapping — Karpicke, Jeffrey D.; Blunt, Janell R. (2011), RCT A doi.org/10.1126/science.1199327 (retrieved 2026-11-09)
7. Distributed practice in verbal recall tasks: A review and quantitative synthesis — Cepeda, N. J.; Pashler, H.; Vul, E.; Wixted, J. T.; Rohrer, D. (2006), meta-analysis A doi.org/10.1037/0033-2909.132.3.354 (retrieved 2026-11-09)
8. Cognitive Offloading — Risko, E. F.; Gilbert, S. J. (2016), systematic review A doi.org/10.1016/j.tics.2016.07.002 (retrieved 2026-11-09)
9. Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips — Sparrow, B.; Liu, J.; Wegner, D. M. (2011), RCT C doi.org/10.1126/science.1207745 (retrieved 2026-11-09)
10. Evaluating the replicability of social science experiments in Nature and Science between 2010 and 2015 — Camerer, C. F.; Dreber, A.; Holzmeister, F.; et al. (2018), RCT A doi.org/10.1038/s41562-018-0399-z (retrieved 2026-11-09)
11. The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers — Lee, Hao-Ping (Hank); Sarkar, Advait; Tankelevitch, Lev; Drosos, Ian; Rintel, Sean; Banks, Richard; Wilson, Nicholas (2025), observational C doi.org/10.1145/3706598.3713778 (retrieved 2026-11-09)