A literature review goes wrong in a predictable way. You start with a question, open thirty tabs, save everything that looks relevant, and two weeks later you have a folder of PDFs and no idea which of them actually answered anything.
The fix is not a better folder structure. It is deciding what the review is for before you collect anything, and then never letting the collection outrun the notes.
1. Write the question down first
Before opening a search box, write one sentence that the review has to answer. Not a topic — a question. “Transformers in biology” is a topic. “Which protein-structure models have been validated on experimental structures released after their training cutoff?” is a question.
The difference matters because a question tells you when to stop. A topic never does.
2. Collect in one place, not in tabs
Browser tabs are not storage. They are a queue that silently drops things when the browser restarts. Put every candidate paper into a single library the moment you see it, with no triage step — triage is a separate activity and mixing it with collection makes you do both badly.
At this stage you are only answering “could this be relevant?”. You are not reading.
3. Take notes against the question, not against the paper
This is the step people skip, and it is the one that decides whether the review converges.
For each paper, do not summarize the paper. Write what the paper says about your question, in one or two sentences, and note where it disagrees with something else you have read. A note that could have been written by someone who never saw your question is not doing any work.
A useful shape:
- What claim does it make that bears on the question?
- What evidence backs that claim?
- Who does it contradict?
4. Follow citations deliberately, twice
There are exactly two directions worth walking, and they answer different things:
- Backwards (what this paper cites) tells you where the idea came from. Use it to find the paper everybody is actually citing when they cite this one.
- Forwards (what cites this paper) tells you whether the result survived. Use it to find the rebuttal, the failed replication, or the version that superseded it.
Doing only the backwards pass is how reviews end up describing a consensus that stopped existing three years ago.
5. Write the synthesis from the notes, not from the papers
If step 3 was done properly, the synthesis is mostly rearranging your own sentences. If you find yourself reopening PDFs to write, that is a signal that the notes were summaries of papers rather than answers to the question — go back and fix the notes, not the draft.
What this looks like in practice
The whole workflow is: one question, one library, notes that argue with each other, two citation passes, then writing. Everything else — tagging schemes, colour codes, folder hierarchies — is optional and mostly decorative.
The only part that is not optional is that the notes have to be written while the paper is still open. A paper you read and did not take notes on is a paper you will read again.