How Far Should Citation Chaining Go Before It Becomes Unproductive?

If Every Relevant Paper Leads to More Papers, When Are You Actually Finished?

Citation chaining has an inconvenient property: it can keep working.

You start with several relevant papers, inspect their references and citing publications, find additional eligible studies, and then realize that those newly discovered studies have their own references and citations. Another iteration produces more candidates. Some of those lead to still more papers.

In principle, a connected scholarly literature can be followed through many generations. In practice, each additional round requires retrieval, deduplication, screening, and evaluation. Eventually the question changes from “Can I find more papers?” to “Is another iteration likely to add enough relevant evidence to justify the work?”

There is no universal number of citation-searching rounds after which every researcher should stop. A defensible stopping decision depends on the purpose of the search, the yield of previous iterations, the sensitivity of other search methods, the structure of the literature, and the methodological requirements of the project.

Stop When Further Iterations Are No Longer Justified by Their Likely Value

Citation Searching Has Diminishing Returns, but No Universal Finish Line

One Citation-Searching Iteration Can Create the Seeds for Another

Suppose your primary database search identifies 15 eligible studies. You use those studies as seed references and perform backward and forward citation searching.

That supplementary search identifies three additional eligible studies.

You now face a methodological choice. The three new studies were not part of the original seed set, and each has its own citation relationships. Searching from them may identify literature that was inaccessible from the first 15 seeds.

Current TARCiS guidance recommends considering another iteration of citation searching when supplementary backward or forward citation searching identifies additional eligible records. The newly included records can therefore become new seed references.

This iterative process is closely related to what researchers often call snowball searching.

An Iteration Means Searching From Newly Identified Relevant Records

The language around snowballing can make iterations sound more mysterious than they are.

Imagine a simple sequence:

The number of iterations is therefore not simply the number of times you click “Cited by.” It refers to repeated citation-searching cycles in which newly identified relevant records become seeds for further searching.

Why Citation Chaining Eventually Produces Diminishing Returns

In a reasonably connected literature, later iterations increasingly encounter papers already reached through earlier searches.

Seed papers may cite many of the same foundational studies. Later publications may cite several of the same influential works. Newly discovered papers may already have been retrieved by the database search, another seed, or the opposite direction of citation searching.

As a result, the proportion of duplicates can rise substantially.

At the same time, the search can move outward from the core research problem. A paper relevant to your question may cite a broader theory, whose references lead to an adjacent field, whose citations lead into another research tradition. The citation connection remains real even as topical relevance becomes progressively weaker.

This combination of increasing duplication and declining relevance is the practical basis of diminishing returns.

Count Unique Relevant Yield, Not the Size of the Citation Network

A citation tool may retrieve thousands of connected records. That number does not tell you whether another iteration was productive.

The more useful question is how many unique relevant or eligible records the iteration added after deduplication and screening.

This hypothetical pattern shows why raw retrieval volume is misleading. The third iteration still found 210 citation-connected records, but nearly all were already known or ineligible.

Whether that is sufficient reason to stop depends on the search objective and methodology, but the marginal yield is much more informative than the size of the result set.

“Stop When You Find No New Papers” Sounds Clearer Than It Really Is

A common informal rule is to continue snowballing until no new relevant papers are found.

That can be a reasonable operational rule in some projects, but it needs clarification.

Does “no new papers” mean no new records before deduplication? No new potentially relevant records at title and abstract screening? No new eligible studies after full-text assessment? No new studies in either citation direction?

These are different stopping points.

For a rigorous search, define what counts as “new” and at what screening stage the stopping condition is assessed.

One Empty Iteration Can Be More Informative Than a Fixed Number of Rounds

Stopping automatically after two iterations is convenient, but the number itself does not reveal whether the citation network has been adequately explored.

Imagine two searches. In the first, the second iteration produces no new eligible studies. In the second, it produces 17.

A rule that stops both searches after two rounds treats very different situations identically.

A yield-based approach can be more responsive to the structure of the literature. If an iteration continues to identify substantial new eligible evidence, another round may be justified. If it produces none, the case for further searching is weaker.

That still does not create a universal rule that every search must continue until one completely empty round. Methodological requirements and available evidence should guide the decision.

Backward and Forward Searching Can Reach Different Parts of the Network

Do not assess productivity as though citation searching were one undifferentiated method.

Backward searching follows references cited by the seed paper. Forward searching identifies later papers that cite it. The two directions expose different portions of the literature.

You may reach a point where backward searching produces only familiar foundational papers while forward searching continues to identify recent eligible studies. Or the reverse may occur for a newer seed with a short forward citation history.

A stopping decision can therefore consider the yield of each direction separately rather than assuming both must remain equally productive.

The Primary Search Changes How Much Citation Searching You Need

Citation chaining should not be evaluated in isolation from the rest of the search strategy.

If a topic is relatively straightforward to express through database terminology and the primary search was designed for high sensitivity, citation searching may function mainly as a supplementary check.

For difficult-to-search topics, citation searching can play a more substantial role. TARCiS recommends seriously considering backward and forward citation searching as supplementary techniques when text-based searching is difficult because of problems such as inconsistent terminology or poor conceptual clarity.

If your database search repeatedly misses eligible studies that citation searching finds, additional citation iterations may have greater value than they would for a topic already retrieved effectively through text-based searching.

The Search Purpose Matters

Not every literature search needs the same stopping standard.

If you are conducting exploratory reading for a research proposal, you may stop once additional citation rounds no longer change your understanding of the field or reveal literature materially relevant to your question.

If you are conducting a systematic review intended to maximize recall, the threshold for stopping should be more demanding and methodologically defensible.

If you are tracing the historical development of a theory, older conceptual predecessors may remain useful even when they would be excluded from an empirical review.

The stopping decision therefore depends partly on what counts as a useful discovery for the project.

Do Not Confuse Search Saturation With Proof of Completeness

Researchers sometimes use the language of saturation when citation searching stops producing new relevant studies.

The metaphor can be useful, but it should not be interpreted as proof that every relevant publication has been found.

Citation networks have boundaries and biases. Relevant work may exist in another discipline, language, database, or weakly connected citation community. Authors can fail to cite related work. Newer publications may not yet have developed substantial citation links.

A Search That Keeps Finding New Papers May Be Drifting

Continued yield is not always evidence that you should continue.

Suppose each iteration finds additional papers, but the papers become progressively less aligned with your original research question. You may be following citation relationships into a neighboring literature rather than improving coverage of your intended evidence base.

This is especially common when a seed paper cites a broad theoretical tradition or multidisciplinary concept.

If one important paper leads into an apparently different literature, pause and evaluate the conceptual relationship before treating that branch as another mandatory iteration.

Scope Creep Can Masquerade as Search Thoroughness

A search becomes unproductive not only when it stops finding papers but also when it starts answering a different question.

You may begin with a narrowly defined intervention and eventually find yourself reading the entire theoretical history of human motivation because one included paper cited a classic theory. The literature is intellectually connected, but that does not make every branch necessary for your review.

Return repeatedly to the research question and eligibility criteria. Citation connectivity is not itself a reason for inclusion.

Deduplication Is Essential for Judging Diminishing Returns

Without deduplication, later iterations can look more productive than they really are.

A new seed may return 150 cited references, but perhaps 130 already appeared in your primary database search or previous citation rounds. Another 15 may be duplicates returned from other seeds in the same iteration.

TARCiS recommends deduplicating citation-search results before eligibility screening and, in iterative searching, deduplicating new results against records retrieved previously.

This allows you to assess the actual marginal contribution of another round.

Search Dates Matter More for Forward Citation Searching

Forward citation networks can continue to grow after your review is completed because new papers can cite existing seeds.

A forward citation search conducted today can therefore produce different results from the same search conducted months later.

For reproducible evidence synthesis, record when the forward search was conducted. TARCiS recommends reporting the date of citation searching when a citation index or similar tool is used.

Manual reference list checking is different because the bibliography of a published seed paper is fixed, but the availability and indexing of those references in external systems can still vary.

There Is No Universal “Correct” Number of Iterations

One round may be enough for one project and inadequate for another.

The TARCiS statement recommends considering an iteration of citation searching when supplementary citation searching identifies additional eligible records, but it does not prescribe a universal number of rounds.

That is appropriate because citation networks differ enormously among topics. A mature, densely connected field behaves differently from a fragmented interdisciplinary literature. A review with two initial included studies has a different starting network from one with 200.

Methodological transparency is therefore more defensible than pretending that a particular number of iterations is universally optimal.

Watching the Yield Decline Across Citation-Searching Iterations

The important feature is not that three iterations are inherently correct. The third round matters because it provides evidence about the marginal yield of this particular citation network.

Another review could reasonably stop earlier or continue longer if its search purpose, citation structure, or results justified doing so.

Common Mistakes When Deciding When Citation Searching Is Finished

Define a Stopping Logic Before the Citation Network Defines Your Project

You do not necessarily need to know the exact number of citation-searching iterations before beginning. You should, however, know what evidence will make another iteration worth considering and what evidence will support stopping.

This turns an otherwise open-ended process into a defensible search procedure.

Track Yield by Iteration

A simple search log can make the stopping decision considerably more transparent.

For each iteration, record the number of seed references, records retrieved, records remaining after deduplication, potentially eligible records, and newly included studies. If you search backward and forward separately, consider tracking their yields separately as well.

You do not need an elaborate dashboard. A small table can reveal whether the search is still opening meaningful new pathways or merely rediscovering the same literature.

Investigate Why Citation Searching Is Still Finding New Studies

If late iterations continue identifying eligible records, ask why your other search methods did not find them.

Perhaps the papers use different terminology. Maybe another discipline is involved. Perhaps database coverage is incomplete. A historical terminology problem may be present.

Those discoveries may justify revising the primary search rather than relying on ever-longer citation chains to compensate for a correctable retrieval weakness.

Distinguish Stopping Citation Chaining From Stopping the Entire Literature Search

You may reasonably stop citation iterations while continuing another search activity.

For example, citation chaining may have reached diminishing returns while an updated database search is still required before publication. A trial-register search may remain incomplete. A newly discovered terminology cluster may justify another targeted database search.

The stopping decision therefore applies to a search method, not automatically to the entire evidence-identification process.

Report What You Actually Did

For reproducible evidence synthesis, vague statements such as “references were snowballed until saturation” leave important questions unanswered.

Current TARCiS guidance recommends reporting the seed references, direction of citation searching, number of iterations, citation index or other method used, search date where applicable, deduplication, screening approach, and other relevant procedures. PRISMA-S likewise provides reporting guidance for citation searching.

A clearer description might state that backward and forward citation searching was conducted from eligible seed records, newly eligible studies were used for subsequent iterations, and searching ended after the specified stopping condition was reached.

The exact wording should reflect what you actually did rather than borrowing a methodological label whose meaning remains ambiguous.

Before Starting Another Citation-Searching Round, Check These Signals

Questions About When to Stop Citation Chaining

Stop Because the Evidence Supports Stopping, Not Because You Reached an Arbitrary Round

How to Cite This Guide

This guide is intended to be read, shared, and used in research, teaching, and academic work. If you draw on its ideas, explanations, or other content, please acknowledge the source by citing the guide. Doing so gives appropriate credit and helps your readers locate the original resource.