How do I organize and extract information from more than 100 papers?

Folder systems fail at scale for a structural reason. A folder forces one paper into one place, but the paper you filed under "methodology" three months ago turns out to matter for your theory chapter, and you will look for it under the theme, not the method. Every research library eventually contains a paper you know you have and cannot find, and the folder tree is why.

Tagging is the usual next move, and it postpones the problem rather than solving it. Tags accumulate faster than they get pruned, near-duplicates appear, and after a year you have "measurement", "measures", and "instruments" holding overlapping sets. Tags work when the vocabulary is fixed in advance, which is exactly what a developing research project cannot do.

The volume is not your imagination. Between 1991 and 2008 the number of Web of Science papers published on malaria rose threefold, on obesity eightfold, and on biodiversity fortyfold (Pautasso, 2013, p. 1). At that rate no filing scheme saves you, so the first thing to fix is intake rather than storage. Record the search terms you used, keep a running list of the PDFs you could not get access to, set your exclusion criteria before you start reading rather than after, and search for existing reviews and not only for primary studies (Pautasso, 2013, p. 1).

What survives is a claim-level record. One row per finding rather than one file per paper. Each row holds the claim, the source, the page, and a note on how strong the evidence is. When your themes shift, and they will, you regroup the rows. Nothing has to be refiled and nothing has to be reread, because the evidence was captured at the level you actually argue with.

Group those rows by argument, not by the order in which the papers arrived. A review that walks through one paper after another with no comparison between them is what Pautasso calls stamp collecting, and it is the standard result of letting the reading order become the structure (Pautasso, 2013, p. 3). Rows let you regroup by shared finding, competing explanation, method, or open question, which is what a reader needs in order to see where the field agrees, where it argues, and what is still unsettled. Keep the search record next to the rows as well, since the database, keywords, and time limits you used are part of what makes the corpus defensible, and a recent cutoff date will quietly drop the foundational work if you let it (Pautasso, 2013, p. 3).

This is the same structure as a synthesis matrix, and it is worth building even if you never show it to anyone. The chapter writes itself out of the rows.

What should I record about every paper I keep?

Four things: the question the paper set out to answer, what it found, the page where the finding sits, and one line on why it matters to your project. Skip the general summary. A summary tells you what the paper is about, which the title already did, and not what it does for your argument.

The page number is the field people leave out and regret. Six months later the difference between a usable note and a dead one is whether you can get back to the passage in ten seconds or have to reread twenty pages. Record the printed page from the journal or book, not the position in your PDF viewer, since those diverge as soon as front matter is included.

The relevance line is the second thing worth protecting. Write it in the first person and be blunt: "supports my argument that X", "the best counterexample I have", "cited everywhere but the data are thin". Future you needs to know what past you thought, and that judgment is the part that never survives in a highlight.

Write the note while you read, not afterwards, and put quotation marks around any wording you copy straight from the paper so you can reformulate it later without accidentally passing it off as your own (Pautasso, 2013, p. 2). Get the reference right at the moment you write the row too. Misattribution is almost impossible to catch once a note has been separated from the source, and the same passage advises capturing your interpretations and your ideas for organizing the review alongside the finding itself (Pautasso, 2013, p. 2).

How do I stop the system becoming a second job?

Cap the effort per paper before you start. Most papers deserve four lines, a handful deserve a page, and you decide which is which during triage rather than while reading. If a note takes longer than skimming the paper again would, the system is costing more than it returns and should be cut back.

The failure looks like this. A note template grows fields because each one seemed useful once. Every paper now takes twenty minutes to process. Processing becomes the work, reading slows down, and the backlog grows, which feels like falling behind and produces more templates.

Two rules hold it in check. First, the note is for retrieval, not for reproduction. It only has to get you back to the paper, and the paper is still there. Second, delete fields you have never once searched on. Almost every heavy note system contains three fields nobody has ever queried, and removing them costs nothing.

A reference manager takes the clerical work off you, and that is all it does. EndNote, Zotero, and Mendeley handle citation organization, bibliography formatting, and collaboration, but a comparison of these tools against newer AI-based ones reports persistent problems with usability, accessibility, and accuracy, and warns that generative systems will produce references that are fabricated or simply wrong (Jin et al., 2026, p. 1). Newer tools such as CiteWell, SemanticCite, and OpenCitations put more weight on reference validation and transparency, but none of them removes the need to check a citation against the paper it points to (Jin et al., 2026, p. 2). Budget time for that check, and budget time for one substantial revision after your rough draft, because a reader coming to the text fresh finds the inaccuracies, ambiguities, and jumps in logic that rereading your own work no longer shows you (Pautasso, 2013, p. 3).