Citation snowballing: backward and forward searching explained

This guide is for students and researchers doing a literature review, a thesis chapter or a systematic review. It is part of our series on finding and checking sources; the pillar guide on levels of evidence explains how to judge the papers snowballing turns up.

What is citation snowballing?

Citation snowballing is a search method that starts from a set of relevant papers and finds more by following their references (backward) and the papers that cite them (forward), round after round. It is also called citation chaining, citation tracking, citation searching or reference harvesting.

Claes Wohlin's 2014 guidelines, written for systematic literature studies in software engineering and now used across fields, describe the method as an iterative procedure: start set, backward and forward snowballing, screening, and repeat with the newly included papers. [1]

What is the difference between backward and forward snowballing?

Backward snowballing is the process of screening the reference list of each paper in your set for relevant older work. Forward snowballing is the process of finding and screening the papers that cite each paper in your set.

Why use citation snowballing as well as a database search?

Snowballing finds papers that keyword searches miss: papers that use different terms, sit in journals your databases don't index, or have vague titles and abstracts.

The evidence for it is strong in some fields. In an audit of a systematic review of complex evidence on health service innovation, Greenhalgh and Peacock found that 252 of 495 primary sources (51%) were found by snowballing, and only 30% came from the protocol-driven database search. [2] That is one review, not a general rate, but it shows how much a database-only search can miss on broad topics.

The Cochrane Handbook makes checking the reference lists of included studies and relevant systematic reviews a mandatory step for Cochrane reviews (MECIR C30), and notes that citation indexes help find studies similar to a known one, because papers citing or cited by a study tend to share its content. [3]

Is snowball sampling the same as citation snowballing?

No. Snowball sampling is a way of recruiting research participants, where each participant suggests others. When people ask about "snowball sampling in a literature review", they mean citation snowballing. In your methods section, write "backward and forward citation searching" to avoid the confusion.

How do you do citation snowballing? Step by step

Run snowballing in rounds, screening as you go. Here is the procedure.

1. Write your inclusion criteria first. You will screen every paper you find against them, so they must exist before round one.
2. Build a start set. Choose papers that clearly meet your criteria. Make the set diverse: different author groups, journals, years and, if relevant, countries. A start set built from one research group snowballs into that group's citation circle. Your database search results are a good source of start papers.
3. Backward round. For each paper in the set, read the reference list. Screen each reference by title, then abstract, then full text, exactly as you screen database results. Skip references you have already seen.
4. Forward round. For each paper in the set, open its list of citing papers in a citation index and screen them the same way.
5. Add the included papers to the set and record where each came from (which paper, backward or forward, which round).
6. Repeat steps 3 to 5 for the newly added papers only. Papers already snowballed don't need a second pass.
7. Stop when a round adds no new included papers, or at the limit you set in advance.
8. Log the counts for every round (see the template below).

Tip: for forward snowballing on a heavily cited paper (say 2,000 citing papers), search within its citing articles for your key terms rather than reading all of them. Google Scholar's "Search within citing articles" option and most library indexes allow this. Report that you did it.

How do I find papers that cite a paper?

Use a citation index: a database that records which papers cite which. Free and subscription options:

Each index covers a different set of journals and catches different citations. For the five or six papers your review depends on, check two indexes. Find references (the backward direction) is easier: the reference list is in the paper, and most databases also show it on the record.

PubMed also offers "Similar articles" on each abstract page, computed from words in the title, abstract and MeSH terms. [4] That isn't snowballing, since it doesn't follow citations, but it is a useful extra way to find similar research papers. Our guide on how to search PubMed covers the rest of the interface.

When should you stop snowballing?

Stop when a full round of backward and forward snowballing finds no new papers that meet your criteria. Wohlin's guidelines iterate until no new papers are found. [1]

In practice, returns shrink fast. Illustrative example with invented figures:

Round 3 added nothing, so the snowballing stops. For a student essay or rapid review you may set a limit in advance ("one backward and one forward round") instead of running to zero. That is a legitimate choice if you state it in your methods.

What are the weaknesses of citation snowballing?

Snowballing inherits the biases of how people cite. Plan for four problems:

• The start set steers everything. A narrow start set gives a narrow result. Fix: build it from your database search, not from memory.
• Citation circles. Groups cite each other, and you can end up with one school of thought. Fix: search deliberately for critics and competing groups.
• Popular papers get more citations. Prominent studies, often the ones with striking results, are cited more, so a snowball can over-represent them. Fix: combine snowballing with a database search, never replace it.
• New papers are invisible to forward snowballing until someone cites them. Fix: run a date-limited database search for the last one to two years.

How do you report citation snowballing in a methods section?

Report the start set, the direction, the indexes, the number of rounds, the stopping rule and the counts. Two reporting guidelines cover it:

• PRISMA-S, the search reporting extension, has an item on citation searching: say whether you examined cited references, citing references or both, and how you located them. [5]
• The PRISMA 2020 flow diagram has a separate column for studies identified via other methods, which includes citation searching, with its own records and reports counts. [6]

Keep a log while you work:

Snowballing log
Start set: 8 papers (list with DOIs)       Inclusion criteria: v2, 3 March 2026
Round | Direction | Source paper (DOI) | Index used | Date | Screened | Included | DOIs included
1     | Backward  | 10.xxxx/...        | Reference list | 4 Mar | 34 | 2 | ...
1     | Forward   | 10.xxxx/...        | OpenAlex       | 4 Mar | 51 | 1 | ...

A sample methods paragraph, to adapt:

We ran backward and forward citation searching on the 8 studies included from the database search. Reference lists were screened from the full texts; citing articles were identified in OpenAlex and Europe PMC on 4 March 2026. Newly included studies were snowballed in turn until a round identified no further eligible studies (3 rounds). Citation searching identified 888 records, of which 14 studies were included.

How to do citation snowballing in AI for Research

AI for Research has a Citation Map built for this method. Searching, mapping and exporting cost 0 credits.

• Citation Map shows the references and citing papers of a seed paper as a graph (years across) and a list, from OpenAlex with Semantic Scholar as a fallback.
• Snowballing rounds: add papers to a set, then expand the next round from the whole set. Candidates are ranked by how many papers in your set link to them, so papers cited by or citing several of your studies rise to the top.
• Reviews and meta-analyses citing your set are listed separately, which is a quick check for a review you have missed.
• Export the set as RIS or CSV for your reference manager, or upload the RIS file into the Systematic Review workspace for screening with your other records.
• Similar papers in the Thesis Suite suggests related papers from a seed (0 credits). It finds papers by similarity, not by citation links, so report it separately from snowballing.

Limits: OpenAlex and Semantic Scholar miss some citations, so check key papers in a second index. The Citation Map doesn't screen for you or keep a round-by-round log; keep the log above. In the Systematic Review workspace's PRISMA diagram, records from other sources are currently counted with the database records, so adjust the diagram if you report citation searching separately.

FAQ

What is citation snowballing?

Citation snowballing is a way of finding literature by following citations from papers you already have. Backward snowballing checks each paper's reference list for older relevant work. Forward snowballing finds newer papers that cite it. You screen what you find, add the relevant papers to your set, and repeat from them until a round turns up nothing new.

What is the difference between backward and forward snowballing?

Backward snowballing looks at the papers a study cites, so it moves back in time and finds the work it built on. Forward snowballing looks at the papers that later cited the study, so it moves forward in time and finds follow-ups, replications and critiques. Do both: backward alone misses everything published after your newest paper.

What is snowball sampling in a literature review?

In a literature review, the snowball method means citation snowballing: following references and citing papers outward from a start set of relevant studies. It shares the name with snowball sampling in field research, where participants recruit other participants, but the two are different methods. In a methods section, write citation searching or backward and forward citation snowballing to avoid confusion.

How do I find papers that cite a paper?

Use a citation index. Free options include the Cited by section on PubMed abstract pages, citing-article lists in Europe PMC, Semantic Scholar and OpenAlex, and the Cited by link in Google Scholar. Scopus and Web of Science have citing-article counts if your library subscribes. Each index covers different journals, so check two for key papers.

When should I stop snowballing?

Stop when a full round of backward and forward snowballing finds no new papers that meet your inclusion criteria. That is the stopping rule in Wohlin's snowballing guidelines. For a rapid or student review, you can set a limit in advance, such as two rounds, and report it. Write the rule in your methods before you start.

Next steps

• Judge the papers you collect: levels of evidence and study types.
• Build the database search that gives you a start set: how to search PubMed.
• Compare graph tools for snowballing: Connected Papers vs. ResearchRabbit vs. Litmaps.
• Keep the growing set in order: how to organize research papers.

Sources

1. Wohlin, C. (2014). Guidelines for snowballing in systematic literature studies and a replication in software engineering. Proceedings of the 18th International Conference on Evaluation and Assessment in Software Engineering (EASE '14), Article 38, 1–10. https://doi.org/10.1145/2601248.2601268
2. Greenhalgh, T., & Peacock, R. (2005). Effectiveness and efficiency of search methods in systematic reviews of complex evidence: audit of primary sources. BMJ, 331(7524), 1064–1065. https://doi.org/10.1136/bmj.38636.593461.68
3. Lefebvre, C., Glanville, J., Briscoe, S., et al. (2025). Chapter 4: Searching for and selecting studies. In J. P. T. Higgins et al. (Eds.), Cochrane Handbook for Systematic Reviews of Interventions (version 6.5.1). Cochrane. https://www.cochrane.org/authors/handbooks-and-manuals/handbook/current/chapter-04
4. National Library of Medicine. PubMed User Guide. https://pubmed.ncbi.nlm.nih.gov/help/
5. Rethlefsen, M. L., Kirtley, S., Waffenschmidt, S., Ayala, A. P., Moher, D., Page, M. J., Koffel, J. B., & PRISMA-S Group. (2021). PRISMA-S: an extension to the PRISMA Statement for Reporting Literature Searches in Systematic Reviews. Systematic Reviews, 10, 39. https://doi.org/10.1186/s13643-020-01542-z
6. Page, M. J., McKenzie, J. E., Bossuyt, P. M., et al. (2021). The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ, 372, n71. https://doi.org/10.1136/bmj.n71 ; flow diagram templates: https://www.prisma-statement.org/prisma-2020-flow-diagram