You know the note exists. You remember roughly when, roughly what it was about, and possibly where you were sitting. You search, and nothing comes back. This is not a failure of memory or of discipline—it is a well-documented property of how people name things, and it has a name and a measurement attached to it.
The vocabulary problem, measured in 1987
George Furnas and colleagues at Bell Communications Research published a paper called ‘The vocabulary problem in human-system communication’ in Communications of the ACM in 1987. Across several experiments they found that two people spontaneously chose the same term for the same familiar object less than twenty percent of the time. The most popular name for a thing was typically picked by only a minority of participants.
The finding matters here because searching your own archive is a version of the same experiment, with you on both ends and months of separation in between. The person who wrote ‘pipeline is stalling’ in March and the person searching ‘sales problems Q1’ in August do not share a vocabulary, even though they are the same person.
Why keyword search cannot rescue you
Traditional search matches the letters you type against the letters that were stored. It handles simple variations well—plurals, capitalization, sometimes stems—and fails completely at synonyms, paraphrases, and descriptions of things that were never named.
The three cases where it fails hardest are exactly the common ones: a note that describes a situation without labelling it, a voice memo where you used casual wording, and a photo whose meaning lives entirely outside its pixels.
Four reasons a note stays hidden
In practice, failed searches come down to a short list of causes. Diagnosing which one you are hitting saves a lot of pointless scrolling.
- Wording mismatch: the note exists and shares no words with your query. Most common by a wide margin.
- Missing context: the note says ‘ask him about it before Friday’ and contains nothing to match on—no person, no topic, no reason.
- Wrong container: it was captured in another app entirely, so no search of this archive can ever return it.
- Never captured: you remember thinking it, not writing it. No system recovers this one, and it is worth ruling out early.
The fix on the retrieval side: search by meaning
Semantic search compares the meaning of your query with the meaning of the stored text, so ‘vendor reliability problem’ can match ‘pipeline is stalling because the supplier keeps slipping’ with no shared words at all. This addresses the vocabulary problem directly rather than working around it.
It works best when you describe rather than label. ‘The conversation where someone objected to the price of the middle tier’ retrieves far better than ‘pricing’, because a longer description gives more meaning to match against. Adding a rough time window narrows the rest.
The fix on the capture side: one line of context
Retrieval cannot repair what capture omitted. A note with no subject, no name, and no reason has nothing to match against no matter how good the search is—the problem is an absence of information, not a failure of technique.
The cheapest habit that fixes this is one closing sentence naming the person, the topic, and why you are saving it. Ten words at capture time replaces ten minutes of searching later. Adding it soon after capture is nearly as good; adding it six weeks later is usually impossible, because by then you no longer know what the note meant either.
What will still be hard
Some questions resist retrieval structurally. Counting—‘how many times did I mention this’—returns an estimate, not a total. Negation—‘which clients did I not follow up with’—asks about things that are absent from the archive, and an archive can only speak to what it contains.
Search quality also degrades with short notes because there is little meaning to work with, and no system can search material you kept somewhere else. If two differently phrased searches fail, ask ‘did I capture this, and where?’ instead of trying more synonyms.
When a search fails, change the query from a label to a description and add a rough date—that fixes most cases. When it fails twice, assume the note lacked context at capture time, and spend the next ten seconds writing the sentence that would have saved you.