I have a folder of highlights going back four years. Articles, textbook chapters, clinical guidance, half-finished books. Until recently I had never once opened it to answer a question. That's the dirty secret of note-taking: most people are running a very expensive write-only system.
AI changes the economics of that, but not in the way the productivity crowd suggests. You don't need a new app. You need to stop organising and start querying. Here's the setup that finally made four years of reading useful.
Why Your Notes Are Currently Dead Weight
Traditional note systems fail at the retrieval step. You file something under a tag that made sense in March, and by September you've forgotten the tag exists. Folders assume you'll remember where you put things. You won't.
A search-based system fails differently: it only finds the words you actually wrote down. If you highlighted a paragraph about spacing effects in memory but later search for "how to make revision stick," you get nothing. The idea was in there. The keyword wasn't.
AI fixes precisely this gap. It matches on meaning rather than string, which means your past reading finally becomes answerable rather than merely storable.
Step 1: Capture Ugly, and Capture Everything
Stop tidying at the point of capture. Every minute you spend formatting a note is a minute you'll never get back, and it doesn't improve retrieval at all now that a model is doing the reading.
What I do: one plain text file per month. Every time I read something worth keeping, I paste the quote, the source, the date, and one line of my own reaction. That last part matters more than anything — your reaction is the bit no article contains and no AI can reconstruct. Thirty seconds, no folders, no tags.
The file grows to maybe 4,000 words a month. That is well within what any current model can hold in a single conversation, which is the whole trick.
Step 2: Run a Weekly Synthesis Pass
Once a week, paste the month's file in and ask for connections rather than summaries. Summaries are useless — you already read the thing. Connections are what a second brain is actually for:
Point three is the valuable one. Contradictions between sources are where you actually learn something, and they're almost impossible to spot manually across weeks of reading. I've caught myself holding two incompatible beliefs about study technique this way more than once.
Step 3: Make It Answer Questions, Not Store Files
The real test of a second brain is whether you can interrogate it. Once a month I dump the whole quarter's notes into a fresh chat and open with a framing instruction:
That last sentence is the whole thing. Without it the model quietly blends your notes with its training data and you can no longer tell which ideas are yours. With it, you get a genuine retrieval tool — and an honest map of the gaps in what you've read.
Step 4: Delete More Than You Keep
A second brain that keeps everything becomes as unsearchable as the internet it came from. At the end of each quarter I archive the raw file and keep only the outputs: the connections, the contradictions, and anything I've since acted on. That's usually about 15% of what I captured.
The discipline here is admitting that most of what you read was interesting rather than useful, and that those are different things. If a note hasn't earned a place in a synthesis in three months, it wasn't going to.
Step 5: Point It at Something Real
None of this pays for itself unless the output goes somewhere. Mine feeds three things: lesson explanations I use with students, articles I write here, and the occasional client project. Reading without an output is a hobby, which is fine — just don't build infrastructure for it.
For most people the natural output is a website, a newsletter, or a portfolio: somewhere your accumulated thinking is visible to people who might hire you for it. If that's the missing piece, it's exactly what I build for people through my web design and build service. A second brain with no shop window is just a very sophisticated diary.
The 10-Minute Loop
To be concrete: 30 seconds per capture as you read, ten minutes on Sunday for the synthesis pass, thirty minutes at quarter-end to prune. That's the entire system. No app subscription, no elaborate tagging taxonomy, no migration weekend.
Start with the last thing you read properly. Paste it into a plain file with one line of your own reaction underneath. That single line — the bit that's yours — is what turns a pile of highlights into something worth keeping.