How to Check Your References for Retractions Before You Cite Them

This is the part you can only see from inside a lot of manuscripts, and it changes how you hunt. In our pre-submission review work with manuscripts, a retracted citation almost never arrives fresh. Nobody reads a retraction notice and cites the paper anyway. It rides in one of three quiet ways, and each is a specific failure pattern worth learning to spot.

The first is the inherited citation. You cite a review or a methods paper, and you carry over its reference to an original study you never actually read. If that original study was later retracted, you have now inherited the flag without ever touching the bad paper. This is why review articles score so high, and why a single sloppy secondary source can seed the same retracted reference across a whole subfield.

The second is the sleeper. The paper was clean the day you saved it, three or four years ago, and it was retracted somewhere in the long gap between your literature review and your submission. Long projects are the ones most exposed. A thesis chapter written in 2023 and submitted as a paper in 2026 has had three years for its citations to go bad in the background.

The third is the batch import. You run a database search or a citation-chasing tool, pull in forty references at once, and format them straight into the draft. Some you skim, most you never open. A retraction anywhere in that batch rides in untouched, because bulk is the enemy of the one-by-one attention that would have caught it.

What we see, again and again, is that all three share one feature: nothing in your workflow surfaces them. You did nothing careless. The reference simply rotted while you were looking elsewhere. That is the whole case for a deliberate check, and it is why we treat it as a standard step rather than a special precaution.

Not every flag is a crisis, and treating them all the same is its own mistake. Three categories, in rough order of how much they should worry you.

Our study found that corrections dominated the raw count, which is the good news hiding in the 7.1%. Most of the flags are corrections, not withdrawals. The 1-in-73 figure is the one that keeps people up at night, and it is the one worth hunting specifically.

There are three ways to do this, and they trade off coverage, effort, and cost. Pick by how many references you have and how much time is left.

By hand, one reference at a time

The manual method works and costs nothing. For each reference you are unsure about, search the Retraction Watch Database by title or DOI. It holds tens of thousands of records. Cross-check anything ambiguous against the publisher's own page, since the retraction notice lives there too.

This is fine for a paper with fifteen references. It falls apart at eighty. Nobody checks eighty DOIs by hand the week before a deadline, which is exactly why the problem survives.

Semi-automated, inside your tools

Some reference managers and browser tools now surface retraction status. Zotero can flag known retractions in your library if you keep it synced. The scite browser extension marks retracted papers as you read. These help, but they are partial: they catch what they happen to know about, and coverage varies by field and by how recently the retraction was logged.

The whole list in one pass

The fast version is to match every DOI in your reference list against the retraction database at once and get back a labeled report. That is what our reference integrity checker does. Paste a reference list or upload the manuscript, and it returns each flagged citation with its category, so you spend your time on the three that matter instead of the eighty that are clean. It is free to run.

The point is not the specific tool. The point is that a one-pass check turns an eighty-reference chore you will skip into a two-minute step you will actually do.

A flag is not automatically a problem. Use these two cuts to decide fast.

Fix it before you submit if: the flag is a retraction or an expression of concern; the citation is load-bearing, meaning your argument or method rests on that paper's result; or the paper appears in your reference list more than once, which usually means it is doing real work in the text.

It is probably fine if: the flag is a correction and you update the citation to the corrected version; the paper is one of several citations backing a routine background sentence and swapping it is trivial; or you are citing the retracted paper on purpose, as the example of the discredited claim, and you say so in the text.

The one option that is never fine is leaving a retraction unlabeled and hoping nobody notices. A reviewer who spots an unflagged retracted citation stops trusting the rest of your reference list, and that suspicion is hard to earn back inside one review cycle.

The 7.1% figure is ours, and here is how we produced it. We built it by pulling the full reference lists of 12,402 recent papers through Crossref, resolving 435,108 unique cited DOIs, and matching each one against the Retraction Watch database. The method is reproducible and the code is published, so anyone can check our work. The full study and per-journal breakdown is public, and the dataset is downloadable under a CC BY 4.0 license. It is a conservative lower bound, since it only counts retractions that were already logged at the time of the check.

That measurement is also why we built the checker. A flagged reference is one of the few problems that is both common and completely preventable, and it is the kind of thing an author is grateful to catch quietly at their desk rather than in a reviewer's report. If you want the reference check as part of a full read of the manuscript, our pre-submission review runs it alongside the rest. For this one problem on its own, the free reference integrity check is all you need.