You've been there. You remember taking a note, maybe it was about a conversation with a client, an idea you had on a flight, a summary of a book you read six months ago. You remember the gist of it. You just can't remember the exact words you used.
So you search. And get nothing.
This isn't a storage problem. It's a search problem. And it's been hiding in plain sight for decades.
Keyword search is pattern matching, not understanding
When you press Ctrl+F or type into a search bar, what happens under the hood is simpler than most people realise: the computer scans your documents for the exact sequence of characters you typed.
That's it.
Type "quarterly review" and it will find every document containing the string "quarterly review." It will miss "Q4 discussion," "end-of-year recap," and "the budget meeting with Sarah", even if all of them describe the exact same thing.
Keyword search has no idea what words mean. It only knows what they look like. It's a very fast, very literal scanner.
This works brilliantly for code ("find every place I called this function"), for databases ("find every row where status equals active"), and for anything where the vocabulary is controlled and consistent. Engineers designed it for machines talking to other machines.
It was never really designed for humans searching through their own thoughts.
Your brain doesn't work that way
Human memory is associative. You don't remember your notes by their exact wording, you remember them by what they were about, how they felt, what context they belonged to, what other ideas they connected to.
When you try to recall something, you rarely reconstruct the exact phrase you originally used. You approach it from a different angle. You think: "that thing about reducing churn... the conversation about retention... the notes from the product meeting in March." Three different phrasings, one memory.
This is completely natural. Language is flexible. The same idea can be expressed dozens of ways, and your brain navigates all of them effortlessly.
Keyword search cannot. Every search is a fresh gamble on whether you happened to use the same words twice.
The vocabulary mismatch problem
The gap widens the more time passes. Notes you wrote six months ago were written by a slightly different version of you, one who might have used different terminology, different abbreviations, different shorthand.
Your present self searches with present words. Your past self wrote in past words. Keyword search sits helplessly in the middle, matching neither.
This is why people give up on personal knowledge bases. Not because they stopped taking notes, but because retrieval became unreliable. The notes are there. The search just can't surface them.
What semantic search does instead
Semantic search approaches the problem differently. Rather than matching characters, it compares meaning.
Here's the intuition: imagine every piece of text, every sentence, every paragraph, every document chunk, being placed on a vast map, where things that mean similar things are placed close together, and things that mean different things are placed far apart.
"Q4 budget discussion" and "end-of-year financial review" would sit very close to each other on this map, even though they share no words. "Chocolate cake recipe" would be far away from both.
When you search, your query gets placed on the same map. The results aren't the documents that contain your exact words, they're the documents that live in the same neighbourhood of meaning.
This is what makes semantic search feel almost uncanny the first time you use it. You type a rough description of what you half-remember, and it surfaces exactly what you were looking for. Not because it got lucky with a keyword, but because it actually understood what you were asking.
Why this matters for personal knowledge
A personal document library is the hardest possible test for search. It's full of prose written in natural language, over months or years, by a you who didn't know what future-you would be looking for. The vocabulary is inconsistent. The topics overlap. The same idea appears in ten different files phrased ten different ways.
Keyword search was built for structured data. Semantic search was built for this.
It's the difference between a library that can only find books if you know the exact title, and a librarian who listens to what you're trying to understand and walks you to the right shelf.
Next: how an AI model actually converts your documents into meaning, and how that powers every search ThinkableSpace runs.