Checking Citations for Retractions Before Journal Submission: A 2026 Guide

Citing a retracted paper is one of the more embarrassing things that can happen to a careful researcher, and it happens constantly. Studies tracking post-retraction citation rates have repeatedly found that retracted papers continue to accumulate citations for years after their withdrawal, often in papers that make no mention of the retraction. In some high-profile cases, the retracted study was cited more times after retraction than before. The reason is straightforward: researchers build reference lists from memory, from previous papers, and from citation manager exports that were built before the retraction occurred. Without a deliberate check, there is no natural point in the workflow where retracted papers get caught.

In 2026, this problem sits at the intersection of two converging developments. First, the volume of retractions in biomedical literature is larger than it has ever been, with biomedical paper retractions having roughly quadrupled over the past two decades. Second, many researchers now build reference lists partly with the help of AI writing tools, and a study published in the Journal of Medical Internet Research in 2026 found that commonly available generative AI tools are not able to detect or appropriately flag retracted scientific literature when asked to assist with citation tasks. Taken together, this means the risk of inadvertently citing a retracted paper is real, the AI assistance many researchers rely on does not protect against it, and the editors who receive your manuscript increasingly have automated tools to catch what you missed.

Why Retracted Papers Keep Getting Cited

Retractions are not always well publicized. When a journal retracts a paper, it typically places a notice on its own website and in PubMed, but older PDF versions of the original paper circulate indefinitely, PDFs downloaded before retraction carry no notice, and secondary sources such as review articles and textbooks that cited the paper before retraction are not automatically updated. The retraction notice and the original article effectively coexist, but researchers accessing only the article encounter no warning.

Citation managers compound the problem. A researcher who added a paper to Zotero or Mendeley in 2022 and has been citing it from that export ever since will not automatically receive an alert when the paper is retracted in 2025 unless they have configured the tool to check for updates, which most do not do by default. The same is true for reference lists assembled from literature review searches. Once the papers are in the document, they tend to stay there unless someone specifically verifies their current status.

The category of retraction also matters. A paper retracted for data fabrication is categorically different from one retracted for an honest duplicated figure, or for a processing error that does not invalidate the conclusions. But most reference list checks do not discriminate. From a submission standpoint, what matters first is whether the paper appears on the Retraction Watch database or carries a retraction flag in Crossref, regardless of the reason. You can make the judgment about whether the finding you cited was affected once you have found the problem, not before.

What the Crossref and Retraction Watch Integration Changed

In September 2023, Crossref acquired the Retraction Watch database from the Center for Scientific Integrity and committed to making it a fully open, continuously updated public resource. Retraction Watch had been the most comprehensive tracker of retractions in scholarly publishing for over a decade, but its data had been proprietary. The Crossref acquisition changed the infrastructure significantly. Data from Retraction Watch is now available through the Crossref REST API, updated on every working day, and accessible without a subscription.

For researchers, the practical consequence is that tools built on Crossref metadata can now incorporate retraction status as a first-class field. Publishers can check submitted reference lists against the database at intake. Reference managers that query Crossref can flag retracted items. Citation checking platforms have a shared, authoritative data source to work from instead of maintaining their own proprietary lists. Not all of these integrations are uniformly implemented yet, but the underlying infrastructure is now in place in a way that it was not three years ago.

The Crossref Retraction Watch database is publicly queryable. For any DOI you want to verify, you can pass it to the Crossref API at api.crossref.org/works/[DOI] and look for a "update-to" or "relation" field that references a retraction or correction notice. This is manual and requires some technical comfort, but for high-stakes citations in a systematic review or clinical trial report, individual DOI lookup through the Crossref API is a reasonable last-resort verification step for papers where automated tools return conflicting signals.

Why AI Writing Tools Are Not the Solution Here

The JMIR 2026 study on AI tools and retracted literature tested a range of commonly used generative AI tools on their ability to identify whether cited papers had been retracted. The finding was direct: freely available AI tools cannot reliably detect retracted scientific literature, and the widespread reproduction of retracted studies through AI-assisted writing represents a distinct threat to research integrity, separate from the fabricated citation problem.

This matters because the two failures are different in character. The fabricated citation problem, which the Lancet research letter in May 2026 documented extensively, involves AI tools inventing citations that do not exist. The retraction problem involves AI tools confidently reproducing citations that do exist but should no longer be used as supporting evidence. A hallucinated citation fails a simple lookup. A retracted citation passes the lookup but fails the retraction check. Researchers who use AI tools for literature synthesis and assume that passing the existence check is sufficient protection are exposed to both failure modes.

Purpose-built scholarly AI tools are better positioned than general-purpose chatbots for this task, but even within the specialized category, coverage varies. Tools trained primarily on publication metadata may not have current retraction data. Tools that do query live databases have to navigate the fact that retraction notices are not consistently registered by all publishers at the same speed. Some retractions appear in Crossref within days. Others lag by months. The Retraction Watch acquisition by Crossref is meant to address exactly this consistency problem over time, but the gap has not fully closed yet.

Tools That Are Worth Using

Several tools now offer retraction checking as part of a broader citation integrity workflow, and it is worth understanding what each one actually does rather than treating them as interchangeable.

Scite.ai is the most widely discussed in this category. Its core feature is Smart Citations, which classifies each citing statement as supporting, contrasting, or mentioning the original claim. The reference check function lets you upload a manuscript or paste a reference list and will flag retracted papers as well as papers that have accumulated predominantly contrasting citations, which is a different but related signal. Scite queries the Crossref/Retraction Watch data and its own corpus of indexed papers. For clinical researchers, scite is particularly useful for understanding not just whether a paper exists but whether subsequent literature has supported or challenged its conclusions. It operates on a subscription model, though some institutions provide access and limited free queries are available.

Paperpal, developed by Cactus Communications, recently added a citation integrity check that includes retraction detection alongside AI hallucination scanning and journal quality review. It pulls from a large article database and cross-references Crossref retraction data. For authors already using Paperpal for language editing, the citation check is integrated into the same interface, which reduces friction.

For Zotero users, the retraction checking situation has improved. Zotero can query the Crossref API for metadata updates, and the Retraction Scanner plugin (a community extension) directly checks library items against Retraction Watch data. The plugin displays a warning badge on flagged items in the library view and during citation insertion in Word. Setup takes about ten minutes, and for users who maintain a large running Zotero library, the scan at manuscript completion takes only a few seconds.

PubMed is also useful and often overlooked for this purpose. Every retracted publication indexed in MEDLINE carries a "Retracted Publication" publication type tag, and searching for your DOI or PMID directly in PubMed will display the retraction notice if one exists. For older papers published before DOI was universal, the PubMed retraction flag may be more reliable than Crossref. If you are submitting to a journal that indexes through MEDLINE, verifying via PubMed is also the closest approximation to what a medical librarian reviewer would use.

EndNote added retraction alerts a few years ago through a Clarivate-built integration. If you are a current EndNote user connected to an institutional Web of Science license, this alert system can surface retracted papers in your library automatically and will flag them during citation insertion. The coverage depends on how quickly Clarivate updates its retraction data, which historically lags somewhat behind Retraction Watch. But it is better than nothing, and for researchers at institutions with Web of Science subscriptions, it requires no additional setup.

A Pre-Submission Reference Check Workflow

The most reliable approach combines an automated scan with targeted manual verification for the citations that carry the most weight in your argument. Not every reference in a 60-item list needs the same level of scrutiny. The citations in your introduction and discussion that directly support your core claims, clinical recommendations, or stated rationale for the study are the ones where a missed retraction would cause the most damage to your paper's integrity.

Start with an automated scan. Export your final reference list from your citation manager and upload it to scite.ai or run it through Paperpal's citation check. If you use Zotero with the Retraction Scanner plugin, run the scan from your library directly. The output will flag any known retractions. Review every flagged item, not just skim the count: a flagged item from a paper retracted for administrative reasons (duplicate submission, licensing error) is different from one retracted for misconduct or data fabrication, and you will want to note the reason as you decide how to handle it.

After the automated scan, take your ten to fifteen most heavily used citations and verify them in PubMed directly. Search by PMID or DOI. Look for the "Retracted Publication" tag and for any correction or expression of concern notices below the abstract. This takes about five minutes and covers the citations where the risk is highest. For clinical trial data, guideline citations, or foundational methodology papers, also check the Crossref API or the Retraction Watch website directly by title, since those resources sometimes carry retraction notices before PubMed updates its tags.

If you are working with co-authors who built parts of the reference list independently, ask each co-author to confirm they have checked their sections. The corresponding author is responsible for the whole manuscript, but you cannot verify citations you did not add if you do not know who added them and when.

Systematic Reviews and Meta-Analyses: The Highest Stakes

For systematic reviews and meta-analyses, retraction checking is not optional and should not be a single end-stage step. The problem with including a retracted primary study in a systematic review is that the retracted data gets pooled into your summary effect estimate, and the error propagates into every future paper that cites your review. A secondary analysis of the literature has found that a meaningful proportion of systematic reviews and meta-analyses in medicine contain at least one retracted primary study. The review authors often did not know about the retraction at the time of writing, but the pooled estimate is wrong regardless.

For systematic reviews, the retraction check should happen at three points: at screening, after full-text review, and immediately before submission. The PRISMA 2020 checklist does not explicitly require authors to report a retraction check, but COPE guidance on systematic reviews notes that authors are responsible for the accuracy and integrity of included studies. Some journals, particularly those with strong methodological review processes, will ask whether retraction checking was performed as part of the review protocol. If your protocol is registered in PROSPERO, you can add a note there about your retraction verification step.

The interval between literature search and submission is another gap worth noting. If a long-running systematic review has an 18-month window between the original search and submission, papers that were current at search may have been retracted in the interim. Running a verification pass at submission, not just at search, catches this. Most systematic review authors do not do this, which is why the literature on post-publication retractions found in meta-analyses keeps documenting the problem.

What to Do When You Find a Retracted Citation

Finding a retracted paper in your reference list before submission is a much better problem to have than finding one after. The course of action depends on what the citation was doing in your manuscript.

If the retracted paper was cited only as background context, alongside other non-retracted sources making the same general point, simply remove it and replace it with a current, intact source if one is available. Add a note in your revision log that you identified and removed the retracted citation during pre-submission review. This is the cleanest case.

If the retracted paper was the primary or sole source for a specific claim in your introduction or discussion, you have more work to do. Check the retraction notice carefully. If the retraction was for reasons that do not invalidate the specific finding you cited (administrative error, licensing dispute, duplicated publication), you may still be able to use the finding with an explicit note citing the retraction notice and the reason. Some journals have specific guidance on this, and it is worth consulting COPE's guidance on citing retracted papers before making that call.

If the retracted paper was for data fabrication or misconduct, and the finding it reported was the basis for a claim you make in your paper, you cannot salvage the citation. The claim either needs to be dropped, reframed as uncertain pending replication, or supported with an entirely different source. If you are in the unfortunate position of discovering that a retracted paper for misconduct was foundational to your study design, that needs a broader conversation with your co-authors about the paper's framing and conclusions before anything goes to a journal.

One scenario that catches authors off guard: you may find a paper on your list that carries a correction or expression of concern rather than a full retraction. Neither of those is a retraction, and the paper is still in the literature. But both are editorial signals worth acknowledging, particularly if the concern touches the specific finding you cited. In those cases, acknowledge the concern in your text if the finding is load-bearing for your argument. Reviewers familiar with the original paper will often notice these, and proactive transparency is better than an uncomfortable revision request.

What Journals Are Starting to Do at Submission

The infrastructure changes at Crossref are making it easier for publishers to run automated retraction checks at submission rather than waiting for peer review to surface problems. Some major medical publishers have begun integrating Crossref retraction data into their manuscript management systems. The exact scope varies by publisher and platform, and not all journals that use these systems make their automated checks visible to authors. But the practical implication is that journals running these checks will catch retracted citations that authors missed, and the resulting editorial exchange is more uncomfortable than if you had caught and corrected the problem yourself before submission.

Springer Nature and Wiley have both discussed using Crossref-based integrity checks as part of their submission workflows, though the specific deployment varies by title within each publisher. Elsevier operates its own integrity-checking infrastructure that includes some retraction verification. The BMJ and JAMA networks have librarians involved in systematic review submissions who often verify primary study integrity as part of editorial review. If you are submitting a systematic review to any of these publishers, assume that retracted primary studies will be noticed.

A Note on Citing Retracted Papers Intentionally

There are legitimate reasons to cite a retracted paper explicitly, as a retracted paper. In a review of the literature on a topic where a landmark study was later retracted, acknowledging that the study existed and was subsequently retracted is honest historiography. In a methods paper discussing citation integrity, retracted papers may be examples. In a discussion of how a field went wrong, citing the retracted study with its retraction notice is appropriate context.

What is not acceptable is citing the retracted paper as if the retraction did not happen, using its findings as current evidence without disclosure. If you intend to cite a retracted paper intentionally, make the retraction visible in the citation or in the running text. Some style guides, including adapted APA and Vancouver conventions, have now added guidance on how to format a citation to a retracted paper to include the retraction notice. Check the target journal's reference style guidelines for specifics.

The Practical Upshot

None of this requires extraordinary effort. The tools are available, most are free or accessible through institutional subscriptions, and the time investment for a standard clinical paper reference list is roughly half an hour when done systematically. The reason most researchers do not do it is not laziness but unfamiliarity. Retraction checking has not traditionally been part of the manuscript preparation workflow in the way that statistical review or figure formatting has been. That is changing as publisher systems become more automated and as retraction rates continue to climb.

The JMIR 2026 finding that AI tools cannot fill this gap is worth taking seriously, particularly as more researchers use those tools for literature tasks. A reference list that was assembled with AI assistance and not subsequently verified through a retraction database is a reference list that has not been checked. The existence of the paper in PubMed is not confirmation that it is still considered valid by the journal that published it. Those are two different questions, and the second one requires a different tool to answer.

Add retraction checking to your pre-submission routine before your next manuscript goes out. The workflow in this guide takes under an hour for most papers and is a clean, documentable step that protects both the scientific record and your own reputation as an author who pays attention to citation integrity.

Further Reading