Is a second brain that depends on your memory really a second brain?
A second brain built on keyword search offloads nothing: it demands that you recall the exact words you wrote at the time. It moves the burden of memorising instead of removing it.

A promise and its blind spot
The idea is appealing and correct: human memory is poor at detail and good at meaning. So hand the detail to a machine and keep for yourself what you are good at — understanding, connecting, deciding.
That is the promise of the second brain, and it has produced solid methods, excellent tools and a community that documents its practices.
But there is a blind spot, and it is logical rather than technical.
The circular argument
Let us put it plainly.
You externalise a piece of information because you will not retain it. That is the whole justification for doing so.
Six months pass. To get it back, a keyword engine asks you for the exact words you wrote that day.
Yet those words are part of what you do not retain. They belong to precisely the category of detail you handed over to the machine.
To give something back, the system demands exactly what you entrusted to it because you could not keep it.
This is not a disappointing user experience; it is a contradiction in the design. The second brain does not offload memory: it shifts what has to be memorised. Before, you had to remember the information. Now you have to remember what you called it.
What survives six months
Look at what your memory actually keeps of a text you read a year ago. Almost never a phrasing. Almost always the shape of an idea:
- "someone showed that measuring a thing distorts it"
- "there was an argument that changed my mind about remote work"
- "a study linked lack of sleep to learning"
None of those three sentences contains a word that is likely to appear in the document. The first describes Goodhart's law without naming it. That is the normal case, not the exception.
A system that accepts only keywords is therefore silent exactly when you need it most: when the memory is vague, which is the ordinary condition of any memory more than a few months old.
Why links are not enough
The expected objection: a good second brain does not rely on search but on links. You navigate from note to note.
True, and it is a genuine strength for thinking. But a link comes with a condition: you need a way in. You must already be holding one end of the chain.
When the memory is "something about metrics that turn against you", you are holding neither end. You do not know which note to start navigating from. The graph is there, magnificent, and it has no door.
What a second brain that kept its promise would look like
It would let itself be queried the way you remember: by meaning, approximately, in plain language.
Technically, that means a search that compares meanings rather than strings, with a lexical engine kept alongside for the cases where the exact word matters — a proper noun, a reference, a figure. Both signals merged into a single ranking.
It would also need to know how to say no. A purely semantic engine always answers something: faced with a query that matches nothing, it returns the three least distant documents. An honest system must be able to return zero results — otherwise you never know whether no answer means "it is not in your base" or "the search failed".
And it would produce the record itself. Transcript, summary, key points: if you have to write the note in order to find it again, the second brain is making you work for it.
The question to ask your own base
Just one, and it takes thirty seconds to check.
Search for an idea you are certain you noted. Phrase it as it comes back to you today, not as you wrote it then.
If it surfaces, your system keeps its promise. If not, what you have is a dump: something that accepts everything and returns nothing. That is not a second brain; it is an archive whose index you have lost.
Frequently asked questions
- Should the second brain concept be thrown out?
- No. The ambition — externalising what you cannot hold in your head — is still sound. The flaw is in the execution: it solved storage and left retrieval to an engine that demands the exact words.
- Does a graph of links solve the problem?
- Partly. A link works when you remember one of its two ends. It is useless when all you have is a hazy idea and no way in.
- How do you check whether your own base holds up?
- A simple test: search for an idea you are certain you noted, but phrase it differently from how you wrote it at the time. If nothing comes back, your base stores without returning.
Published 2026-09-04