Every literature review starts the same way: you pick a handful of keywords, run a search, and read what comes back. It works, right up until it doesn't — because keyword search has a blind spot built into it. It can only find papers that happen to use your words. A foundational paper written before your field settled on its current vocabulary, a study from an adjacent discipline that solves your problem under a different name, a niche follow-up that never made it into anyone's abstract keywords — none of that shows up, no matter how many synonyms you try.
Citation snowballing is the fix, and it's a genuinely old technique — researchers have used it for decades, long before "citation graph" was a product feature. The idea is simple: instead of searching by words, you search by relationships. You start with one paper you already trust, and you follow its citations outward.
The two directions
Snowballing has a direction, and it matters which way you go.
Backward snowballing means looking at what your paper cites — its references. This takes you earlier: the foundational work, the methods it borrowed, the studies its argument rests on. If a paper makes a claim you want to verify or build on, its references are where that claim actually comes from.
Forward snowballing means looking at what cites your paper — who's citing it. This takes you later: newer work that builds on it, critiques it, replicates it, or extends it into a new context. Forward citations are how you find out whether a finding held up, and what happened to it after publication.
Do both, and you're not just reading in one direction along a timeline — you're triangulating a paper's whole position in the literature: what it's built on, and what's been built on it since.
Why it finds what keyword search misses
The mechanism is what makes this work. Keyword search matches text. Citation snowballing matches ideas, because a citation is an author telling you, directly, "this other paper is relevant to what I'm doing here" — regardless of whether the two papers share a single word of vocabulary. A 2003 paper on "customer churn prediction" and a 2019 paper on "subscriber attrition modeling" might never turn up in the same keyword search. But if the second cites the first, snowballing connects them instantly, because the connection an author actually made is worth more than a string match.
This is also how you catch the papers that matter most and search the least well for: older foundational work (vocabulary drifts over time, but citations don't), cross-disciplinary work (a psychology paper solving your CS problem under psychology's terminology), and narrow follow-up studies too specific to rank well on their own keywords but very findable once you know which paper they're extending.
Run it a couple of hops deep — snowball from a paper you found by snowballing — and you'll usually surface the small, tight cluster of papers that actually define your sub-topic, faster than any amount of keyword refining gets you there.
Doing it in two clicks
Folio Search builds this in as a standard part of any result. If a result has a DOI, it carries two buttons: Cited by and References.
- Click References to expand every source that paper cites — its own backward snowball, right there inline.
- Click Cited by to expand everything that cites it — the forward snowball, sorted by influence.
Each paper that shows up is a full result in its own right — title, authors, year, citation count — and it saves to your library with one click, same as anything else in search. You can chase a thread three or four papers deep without ever leaving the results page or opening a new tab, and the moment something's worth keeping, it's already in your library instead of a browser tab you'll lose track of.
That's the whole workflow. Find one solid paper — a survey, a paper you already trust, the one everyone in your field cites — and then spend ten minutes walking outward in both directions. It reliably turns up more than another round of keyword tweaking ever does.
Folio Search pairs full-text discovery with a live citation graph on every result, so you can search by keyword and by relationship in the same pass. Try it.