Snowballing in a Literature Review: How to Find Papers by Citation Chaining (2026)

Snowballing, also called citation chaining, means finding new sources by following the citation links of papers you already have instead of running another keyword search. It works in two directions. Backward snowballing reads a paper’s reference list to find the earlier work it was built on. Forward snowballing finds the later papers that have cited it since. You start from a small seed set of central papers, chain in both directions, add anything that meets your criteria to the seed set, and repeat until new rounds stop producing anything new.

It is the technique that most reliably rescues a review whose search felt thorough and whose results felt thin, because it finds papers by relationship rather than by wording. This guide covers both directions in detail, how to choose seeds, which tools do which direction, the bias snowballing introduces if you lean on it too hard, and how to report it so an examiner can repeat it.

Why Citation Chaining Finds What Searching Misses

A keyword search matches strings. A citation matches ideas. That one sentence is the whole argument for snowballing, and it explains the failure mode it fixes: two research communities can study the same phenomenon for a decade under different names, and no amount of care with your search terms will bridge them if you do not know the other name exists. A researcher who worked across both, though, cited both, and their reference list is where the bridge is.

This is also why snowballing tends to improve the quality of a review rather than just its size. A reference list is not a neutral list of related work: it is a record of which papers someone judged worth arguing with. Following it puts you in contact with the debate rather than with a pile of topically similar studies, and a literature review is a conversation you are joining rather than a stack of summaries. Papers found by citation arrive already related to each other, which is exactly the raw material synthesis needs.

Backward Snowballing, Step by Step

Backward snowballing goes into the past: you take a paper you already have and mine its reference list for the work it stands on. It costs nothing and needs no tool, since the list is already in the PDF.

Read the reference list with the paper still open, not on its own. The trick that makes this fast is to work from the in-text citations rather than alphabetically down the back page: find the places in the introduction and the related-work section where the author says something like “this builds on” or “in contrast to”, and chase those citations first. A reference cited once in passing in a methods footnote and a reference the whole argument rests on look identical in the bibliography, and only one of them is worth your afternoon.

For each candidate, screen on title first, then on abstract, against the same criteria you use everywhere else in the review. A snowballed paper gets no easier ride for having been cited by someone you respect. Keep a note of which seed paper it came from, because that record is what lets you reconstruct the trail later and what turns a vague “I also checked reference lists” into a reportable method.

Review articles and meta-analyses are the highest-yield backward seeds by a wide margin. Their reference lists are the output of somebody else’s systematic search, which means one hour spent on a good review’s bibliography routinely beats a day of database work. The catch is the publication date: a review from four years ago captures the field as it stood then, which is precisely why the forward direction exists.

Forward Snowballing, Step by Step

Forward snowballing goes into the future: you take a paper and ask who has cited it since. This is the direction that finds the current state of a debate, including the replication that failed, the critique that reframed the question, and the extension into your specific setting.

You need a citation index for this, because the information does not exist in the paper itself. Google Scholar shows a Cited by link with a count under each result, and clicking it gives you the citing papers as a searchable set, which matters when a landmark paper has thousands of them. Semantic Scholar and OpenAlex both expose citations and references and are free to query. Web of Science and similar subscription indexes offer the same function with different coverage, and your library may already provide access.

When a seed paper is heavily cited, do not scroll. Use the search within citing articles function to filter that set by a term from your own question, or sort by date and take the most recent two years. This is the single highest-value ten minutes in the whole method for a review that needs to look current: the papers published since your search terms were fixed are, almost by definition, the ones your search was least likely to phrase correctly.

Choosing Your Seed Set

Snowballing amplifies whatever you feed it, so the seed set is the decision that determines the result. Three to five papers is enough to start. What matters is that they are central to your question rather than merely relevant, and that they differ from each other: a recent empirical study, a review or meta-analysis, and something from a different tradition, region or discipline that approaches the same problem another way.

Five papers from the same research group will chain into the same neighbourhood five times and leave you confident about a field you have only seen one corner of. Deliberate variety in the seeds is the cheapest available correction, and if your seeds all cite each other, that is the signal to go looking for a paper that disagrees with them before you chain any further.

Which Tool Does Which Direction

As of August 2026 the tools below cover the two directions between them. Coverage differs between citation indexes, so a paper showing 40 citing works in one and 60 in another is normal rather than an error, and checking two is worthwhile for a key paper.

How Many Rounds, and When to Stop

Snowballing is iterative, which is where the name comes from. Papers that pass screening in round one become seeds for round two, and so on. In practice two or three rounds is usually enough for a masters or doctoral literature review, and the returns fall off sharply after that.

The stopping rule is saturation: you stop when a round of chaining returns papers you have already seen. That moment is worth trusting when it arrives from several directions at once, and worth doubting when it arrives early from a narrow seed set, which is the same thing as saying it is only meaningful if your seeds were varied. Formal guidance exists for this if your method needs to cite something: Claes Wohlin’s 2014 paper on guidelines for snowballing in systematic literature studies, published in the proceedings of the EASE conference, is the usual reference point.

Set a time budget as well as a stopping rule. Citation chaining is unusually absorbing, and it is possible to spend a fortnight inside a citation network and emerge with sixty papers and no argument.

The Bias You Inherit, and How to Correct It

Here is the honest limitation, and it is the reason snowballing supplements a search rather than replacing one. Citation networks are not neutral. Highly cited work attracts more citations, which means chaining systematically favours the established centre of a field and systematically under-samples the periphery: recent work that has not accumulated citations yet, research from less well funded institutions and regions, and studies that reported null results and were quietly cited by nobody.

Three corrections cost little. Run a proper database search first, so your seeds are not simply the papers that happened to reach you. Sort at least one forward search by date rather than relevance, which surfaces the newest citing work regardless of how cited it is. And go looking specifically for the tradition your seeds do not cite, which is often where the sharpest disagreement in your review will come from. Grey literature belongs in the same correction, since reports and theses are rarely well cited and are frequently where null results live.

One more check before a snowballed paper goes into your draft: citation networks preserve papers long after the record has moved on, and a retracted study can keep being cited approvingly for years, so it is worth knowing how to check whether a paper has been retracted before you build a paragraph on one.

How to Record and Report It

Snowballing is a search method, so it is described in your methods with the same specificity as a database search. Record the seed papers, the directions you chained, the tools and indexes you used for the forward search, the dates you ran it, the number of rounds, and your stopping rule. Keeping a column in your source table for how each paper was found makes this write itself.

For reporting counts, PRISMA 2020 distinguishes records identified from databases and registers from records identified by other methods, and citation searching sits in that second group with its own path through the diagram. The PRISMA flow diagram guide shows where those boxes sit. Since snowballing runs after your main search rather than instead of it, it is worth reading alongside how to write a search strategy, and how to find academic sources covers where the seed papers come from in the first place.

Frequently Asked Questions

What is snowballing in a literature review?

Snowballing is finding new papers by following the citation links of papers you already have, rather than by running another keyword search. Backward snowballing means going through a paper’s reference list to find the earlier work it built on. Forward snowballing means finding the later papers that have cited it since. Together they are also called citation chaining, and they surface work that uses different vocabulary from your search terms.

Is snowballing a substitute for a database search?

No, it is a complement, and using it alone introduces a specific bias. Citation networks favour work that is already well cited, so a review built only by snowballing tends to inherit the blind spots of its starting papers and to under-represent recent, unfashionable, or differently framed research. The standard approach is a systematic database search first, then snowballing from the papers it returns to catch what the search terms missed.

Which papers should I snowball from?

Choose a seed set of roughly three to five papers that are genuinely central to your question rather than merely relevant, and make them varied: a recent empirical study, a review or meta-analysis, and a paper from a different school of thought or region. A review article is the highest-yield backward seed because its reference list is already a curated search. Snowballing from five near-identical papers mostly returns the same results five times.

What free tools do citation chaining?

Google Scholar’s Cited by link handles forward chaining on a single paper, Semantic Scholar and OpenAlex expose both citations and references and are free to search, citationchaser runs backward and forward chaining on a set of papers at once, and Research Rabbit and Connected Papers show the citation network visually. Reference lists in the PDF itself cost nothing at all and remain the most reliable backward source.

How do I report snowballing in my methods or PRISMA diagram?

PRISMA 2020 separates records found through databases and registers from records identified by other methods, and citation searching belongs in that second group with its own counts. In your methods, say which papers you snowballed from, whether you went backward, forward, or both, which tools and databases you used for the forward search, the date you ran it, and when you stopped. A reader should be able to repeat it.