Search over sort

You spent more time filing it than you will ever spend finding it

You have built the folder structure. Maybe it took an afternoon. Maybe it evolved over months — a new subfolder for each project, a new category each time something did not fit neatly into the existing ones. Your file system now has four levels of hierarchy. Business contains Strategy contains Pricing contains Frameworks. Personal contains Health contains Nutrition contains Recipes contains Desserts. The taxonomy is thorough. The labels are precise. And every time a new piece of information arrives, you spend thirty seconds to two minutes deciding which branch of this tree it belongs on.

Now multiply that filing decision by every piece of information you store in a year. If you save three items per day — a conservative estimate for anyone engaged in knowledge work — that is roughly a thousand filing decisions per year. At an average of forty-five seconds per decision, you have spent twelve hours annually on the organizational equivalent of alphabetizing your bookshelf. Twelve hours deciding where things go.

Here is the question nobody asks: how much time did you spend getting things back?

The previous lesson taught you that time-sensitive information should carry an expiration date so it does not clutter your system indefinitely. This lesson challenges a deeper assumption — that the way to make information findable is to organize it into an increasingly refined hierarchy. It is not. The way to make information findable is to make it searchable. And those are not the same thing.

Pilers and filers: what the research actually shows

In 1983, Thomas Malone published a study at MIT called "How Do People Organize Their Desks?" that became one of the foundational texts in personal information management research. Malone observed that knowledge workers fell into two broad categories in how they handled documents on their physical desks: filers and pilers.

Filers maintained clean desks. Every document had a designated place in a filing cabinet or folder system. When a new document arrived, it was immediately classified and filed. The desk surface stayed clear.

Pilers maintained messy desks. Documents accumulated in stacks — on the desk surface, on chairs, on the floor. There was no formal filing system. When a piler needed a document, they rummaged through the stacks, relying on spatial memory ("it was near the bottom of the pile on the left side of my desk") and temporal memory ("I got it sometime last week, so it should be about this deep in the stack").

The intuitive assumption is that filers performed better. They were organized. They had systems. They invested in structure. But Malone's research found something more nuanced: pilers were not significantly worse at retrieval than filers, and in some cases they were faster — because their documents were physically accessible on the desk surface rather than buried in a filing cabinet. The filer's advantage was aesthetic and psychological (the desk looked better, and they felt more in control), but the retrieval performance difference was smaller than anyone expected.

This finding has been replicated and extended in the digital domain. Richard Boardman and Angela Sasse, in a 2004 study at University College London titled "Stuff Goes into the Computer and Doesn't Come Out," conducted a cross-tool study of personal information management. They found that participants who maintained elaborate folder hierarchies for their files, emails, and bookmarks did not retrieve information faster than participants who relied on simpler structures. In several cases, the heavy filers were slower, because they had to recall which of their many categories they had chosen — a recall task that becomes harder as the number of categories increases.

The consistent finding across decades of research is this: the retrieval benefit of folder hierarchies flattens quickly. The first level of organization — having any system at all versus having none — produces a large improvement. The second level — a few broad categories — adds modest value. Every level beyond that adds filing cost without proportional retrieval benefit. And with the arrival of full-text search in digital tools, the equation has shifted decisively: search is now faster than navigation for nearly every retrieval task.

The cost of sorting that nobody counts

Every folder you create carries three hidden costs that accumulate over time.

The filing decision cost. Each time you save a new item, you must decide where it goes. With three folders, this is trivial. With thirty, it requires genuine thought. With three hundred, it becomes a non-trivial cognitive task — and one that has no right answer, because most information belongs in multiple categories simultaneously. Is an article about the psychology of pricing a Business item or a Psychology item? Is a recipe you found for a client dinner a Cooking item or a Work item? Is a tax-related investment statement a Finance item, a Taxes item, or an Investments item? Each ambiguous decision consumes time and creates a retrieval risk, because if you file it under the "wrong" category — whichever one your future self does not think to check first — it becomes effectively lost.

The maintenance cost. Folder hierarchies are not static. As your information landscape changes, the categories need to change too. Projects end. New topics emerge. Old categories become overcrowded and need to be split. Empty categories accumulate and create visual noise. The hierarchy needs pruning, restructuring, and updating — administrative work that produces no direct value. You are organizing the organizational system rather than using the information within it.

The cognitive overhead cost. When you know you have an elaborate filing system, you feel an obligation to maintain it. A new item arrives, and instead of spending five seconds giving it a clear title and dropping it into a flat pool, you spend ninety seconds navigating the folder tree, debating between two candidate locations, and sometimes creating a new subfolder to resolve the ambiguity. This friction accumulates into a background sense that filing is burdensome — which leads to the most predictable outcome: you stop filing altogether, and information scatters across your inbox, your desktop, your downloads folder, and your browser tabs.

The cruel irony is that the more elaborate your filing system, the more likely you are to abandon it. Simple systems get used. Complex systems get admired briefly and then ignored.

The search revolution you already participated in

Consider how you find information on the internet. You do not navigate a hierarchy. You do not open a folder called "Medical Information," then a subfolder called "Conditions," then a subfolder called "Musculoskeletal," then a subfolder called "Back Pain," then browse a list of documents to find the one about stretches for lower back pain.

You type "stretches for lower back pain" into a search box, and the answer appears in under a second.

Google did not organize the internet into folders. It made the internet searchable. And in doing so, it rendered the question "where does this belong?" irrelevant. What matters is not where information lives in a hierarchy. What matters is whether the right information surfaces when you describe what you need.

This same shift has happened — more quietly, but just as decisively — in personal information management. The tools you already use have search capabilities that most people dramatically underutilize.

Full-text search in note-taking apps. Obsidian, Notion, Apple Notes, Evernote, Google Keep, and nearly every modern note-taking tool can search the full text of every note in your system. You do not need to remember the title or the folder. If the words exist anywhere in the note — in the title, the body, the tags — search will find them. A note titled "Meeting notes 3/14" is hard to find by title alone. But if it contains the sentence "Sarah proposed shifting the launch date to Q3 based on the supply chain delay," you can find it by searching "Sarah launch date" or "supply chain delay" or "Q3."

Desktop search tools. On macOS, Spotlight indexes every file, email, and application on your machine. On Windows, the built-in search and third-party tools like Everything (which indexes file names in milliseconds) make finding any file a matter of typing a few characters. Alfred and Raycast on macOS add layers of intelligence — recent files, file content search, and custom workflows that surface information faster than any folder navigation could.

Email search. Gmail's search, Outlook's search, and comparable tools in other email clients make navigating email folders almost entirely unnecessary. You do not need a folder called "Receipts" if you can search "receipt" and get every receipt ever sent to you. You do not need a folder called "Project Alpha" if you can search "project alpha" and get every email thread related to it. The people who maintain fifty email folders are doing manually what the search engine does automatically — and the search engine does it faster.

The infrastructure for search-first information management already exists. You are already using it for external information (via Google). The shift this lesson asks you to make is to use it for your own information too.

What search needs from you

Search is not magic. It operates on the information you provide, and it fails predictably when that information is poor. Adopting a search-first approach does not mean abandoning all organizational discipline. It means redirecting that discipline from structure to searchability.

Here is what search requires to work well:

Descriptive titles. A note titled "Notes" is unsearchable. A note titled "Vendor evaluation framework — weighted scoring criteria for SaaS procurement" is highly searchable. The title should contain the words your future self will type when they need this information. This is the retrieval-first naming principle from The reference filing system, and it matters even more in a search-first system — because the title is the primary surface that search operates on.

Natural-language content. Search works best when your notes are written in the language you actually think in. If you write in shorthand, abbreviations, and cryptic fragments — "VE frmwk, 3 criteria, wt avg" — search will struggle because you will search in full words ("vendor evaluation framework") that do not match what you wrote. Write your notes in complete enough language that a search for any key concept will match.

Consistent terminology. If you call the same concept "vendor evaluation" in one note, "supplier assessment" in another, and "procurement scoring" in a third, a single search will not find all three. You do not need a formal controlled vocabulary, but developing awareness of your own terminology patterns — and occasionally standardizing — makes search dramatically more reliable.

A few broad tags for browsable categories. This is the middle ground between no organization and excessive organization. Tags are not folders. An item can have multiple tags without being duplicated. Tags serve a different purpose than search: they answer the question "show me everything related to X" rather than "find the specific item about Y." A small set of tags — ten to twenty — provides browsable groupings without the filing-decision overhead of a deep hierarchy. Think of tags as search shortcuts for broad contexts you revisit frequently.

The organizing spectrum: from pure filing to pure search

This lesson is not arguing that all organization is wasteful. It is arguing that most people are dramatically over-investing in filing structure and under-investing in searchability. The optimal approach lies along a spectrum, and your position on it depends on your volume, your tools, and your retrieval patterns.

Level 0: No system. Information scattered across fifteen different locations with no naming conventions and no search habits. Retrieval is essentially random. This is where most people start, and it is genuinely dysfunctional.

Level 1: Single location with search. All information goes into one system. Items have descriptive titles. Search handles retrieval. No folders. No categories. This is surprisingly effective for systems with up to several thousand items, especially with full-text search. For most people, this level provides 80 to 90 percent of the retrieval benefit of any system, at roughly 10 percent of the organizational effort.

Level 2: Single location with search and broad tags. Same as Level 1, plus ten to twenty tags for major categories. Tags allow browsing when you want to see a group of related items rather than find a specific one. The filing decision per item increases by about five seconds (select one or two tags from a short list). The retrieval benefit is modest but real for broad queries.

Level 3: Single location with search, tags, and a few top-level folders. Three to seven broad folders — Projects, Reference, Archive, or the PARA categories from Tiago Forte's system — provide a first layer of structure. Items are filed into one of the few top-level buckets and then found by search within that bucket. This level adds minimal filing cost and helps when the total item count is very large, because search within a scope (search my Archive, search my Projects) reduces noise.

Level 4: Deep folder hierarchy. Multiple levels of nesting, each folder containing subfolders containing sub-subfolders. This is where the cost-benefit ratio inverts. Filing decisions become time-consuming. Retrieval via navigation becomes slower than search. Maintenance becomes a task in itself. The system starts to feel like an end rather than a means.

Most knowledge workers should operate at Level 1 or Level 2. Some, with large volumes and specific browsing needs, benefit from Level 3. Almost nobody benefits from Level 4 — and almost everybody starts there, because it feels like the responsible, organized thing to do.

The desktop metaphor and its quiet failure

The reason most people default to deep folder hierarchies is that the metaphor is built into the operating system itself. Since the Xerox Star in 1981 and the Apple Macintosh in 1984, personal computers have used the "desktop metaphor" — files live in folders, folders live inside other folders, and navigation means drilling into a tree structure. The metaphor was borrowed from physical office furniture: the desktop, the file folder, the filing cabinet.

The metaphor was brilliantly effective for making computers approachable to people who had never used them. It was never meant to be the optimal way to manage information at scale. Physical filing cabinets needed folders because they had no search function — you literally had to browse through folders to find a document. The digital equivalent of a filing cabinet has had full-text search for decades, but the visual metaphor persists, and with it the assumption that finding a file means navigating to it rather than searching for it.

Yukio Noguchi, the Japanese economist who invented the Noguchi Filing System referenced in The reference filing system, recognized this mismatch in the early 1990s — before most people had personal computers. His system eliminated categories entirely. Documents were filed in chronological order along a single shelf. When you needed a document, you pulled it out; when you were done, it went back on the left end of the shelf. The most frequently accessed documents naturally migrated to the left. Rarely used documents drifted to the right. The system required zero categorization decisions and worked because the two strongest retrieval cues for most documents are recency (when did I last use this?) and frequency (how often do I use this?) — not category.

In the digital world, "sort by date modified" is the Noguchi method. And for a surprising number of retrieval tasks, it is all you need.

The real argument: invest in searchability, not in organizational complexity

The key nuance of this lesson is worth stating explicitly, because it is easy to hear "search over sort" as "do not organize at all." That is not the argument.

The argument is about where to invest your organizational energy. You have a finite amount of time and attention for information management. Every minute you spend building folder structures is a minute you did not spend writing better titles, adding useful tags, or writing notes in searchable language. Every minute you spend maintaining a hierarchy is a minute you did not spend processing the information itself — extracting key ideas, making connections, deciding what matters.

Some structure helps search. A few broad categories reduce noise. Tags enable browsing. A clear naming convention makes every search more effective. These investments in searchability have high returns.

Excessive structure hinders everything. Deep hierarchies create filing friction, navigation delays, and maintenance burdens. They solve a problem that search already solves — and they solve it more slowly, at higher cost, with more room for error.

The heuristic is: when in doubt, make the information more searchable rather than more organized. Write a better title. Add a tag. Include key terms in the first paragraph. Then drop it into the simplest structure that your system supports and trust search to do the rest.

Your Third Brain: AI as retrieval intelligence

AI does not just improve search — it transforms the nature of what "searching" means.

Semantic search. Traditional keyword search requires you to match the exact words in the document. If you wrote "weighted scoring criteria for procurement" and later search for "how to compare vendors," traditional search returns nothing — zero keyword overlap. Semantic search, powered by AI embedding models, understands that these phrases refer to the same concept. Tools like Obsidian with the Smart Connections plugin, Notion AI, and Mem use vector embeddings to match meaning rather than words. This is the single most important development in personal information retrieval in the past decade, because it eliminates the most common search failure: the vocabulary mismatch between your filing self and your retrieving self.

Conversational retrieval. Instead of composing a search query, you describe what you are looking for in natural language. "I wrote something a few months ago about how to evaluate vendors — it had a scoring framework with weighted criteria" is not a search query. It is a description. AI can take that description, infer the key concepts, search your note corpus, and return the most relevant matches. This works particularly well for the retrieval cases that traditional search handles worst: when you remember the gist of the information but not the specific terminology you used when you filed it.

Automatic tagging and categorization. If tagging adds retrieval value but you do not want to spend time on tagging decisions, AI can suggest or automatically assign tags based on the content of each note. You write the note, the AI reads it, and it suggests two or three tags from your existing tag vocabulary. You approve with a single click. The searchability benefit of tagging is preserved, but the filing-decision cost drops to near zero.

Cross-note connection surfacing. When you create or revisit a note, AI can scan your entire corpus and suggest related notes — even ones you filed years ago, in different terminology, in different contexts. This is the connective function that the Zettelkasten (The Zettelkasten method) handles through manual linking. AI does not replace the cognitive value of deliberate linking, but it dramatically reduces the chance that a relevant connection goes undiscovered simply because your note volume exceeds what your memory can hold.

The sovereignty principle applies: the AI does not decide what information is worth keeping, how to organize your system, or what your retrieval priorities should be. It makes whatever system you have built more effective at the task search was always designed to do — getting the right information back to you at the moment you need it.

The bridge to progressive summarization

You now have a principle for how to invest your organizational energy: make information searchable rather than sorted. Write clear titles. Use consistent terminology. Rely on a few broad tags and full-text search rather than deep folder hierarchies. Let AI-powered semantic search close the gap between how you describe information at filing time and how you describe it at retrieval time.

But findability is only half the retrieval problem. The other half is density. You find the note you were looking for — and it is 2,000 words long. You need the core insight, not the full text. You need the three sentences that capture the key idea, not the fifteen paragraphs of context and evidence that surround them. You need the information concentrated to the point where finding it and using it are nearly the same action.

The next lesson introduces progressive summarization — a technique for systematically concentrating the value of stored information through multiple passes of highlighting and distillation. If this lesson makes your information findable, the next lesson makes it immediately usable once found.

---

Sources:

1. Malone, T. W. (1983). "How Do People Organize Their Desks? Implications for the Design of Office Information Systems." ACM Transactions on Office Information Systems, 1(1), 99-112.
2. Boardman, R., & Sasse, M. A. (2004). "Stuff goes into the computer and doesn't come out: A cross-tool study of personal information management." Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 583-590.
3. Jones, W. (2007). Keeping Found Things Found: The Study and Practice of Personal Information Management. Morgan Kaufmann.
4. Barreau, D., & Nardi, B. (1995). "Finding and reminding: File organization from the desktop." ACM SIGCHI Bulletin, 27(3), 39-43.
5. Noguchi, Y. (1993). The "Super" Organization Method (Cho Seiri-ho). Chuko Shinsho.
6. Allen, D. (2001). Getting Things Done: The Art of Stress-Free Productivity. Viking.
7. Forte, T. (2022). Building a Second Brain: A Proven Method to Organize Your Digital Life and Unlock Your Creative Potential. Atria Books.
8. Lansdale, M. W. (1988). "The psychology of personal information management." Applied Ergonomics, 19(1), 55-66.
9. Bergman, O., Beyth-Marom, R., & Nachmias, R. (2006). "The project fragmentation problem in personal information management." Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 271-274.
10. Cutrell, E., Robbins, D., Dumais, S., & Sarin, R. (2006). "Fast, flexible filtering with Phlat: Personal search and organization made easy." Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, 261-270.