Why Retracted Papers Keep Getting Cited

A retraction notice tells readers that a paper's findings or integrity can no longer be relied upon in the way originally published. It does not delete the paper from history: the DOI continues to resolve, the bibliographic record remains, and copies may sit on laptops, repositories, course pages, reference managers, and unofficial archives.

That preservation is intentional. A transparent scholarly record needs the original item, a prominent status label, and a linked notice explaining what changed, and the failure occurs when status does not travel with content.

Studies across biomedicine have repeatedly found citations after retraction, often without any acknowledgment that the cited work was withdrawn, though some later citations discuss the retraction as an example, which is appropriate. Others continue to use the findings as valid evidence. Citation counts alone cannot distinguish the two.

Retraction is a status change, not disappearance#

COPE describes retraction as a mechanism for correcting the literature when findings are unreliable or publication integrity is seriously compromised. The recommended practice is to keep the article available, mark it clearly, and link a notice that identifies the reason and who is retracting it.

Removing the record would create other problems. Readers could not audit how later work was affected, distinguish a missing link from a withdrawn article, or study research integrity. Legal and historical accountability would weaken. The durable record depends on visible version control. A bare PDF downloaded before the notice may contain no live status check and can circulate as though nothing happened.

Retraction often arrives years later#

Investigations take time. Concerns may emerge after image comparison, attempted replication, data-access requests, whistleblower reports, or institutional review. Journals seek responses and may wait for institutions, funders, or courts.

During that interval, the article can accumulate hundreds of citations and enter reviews, teaching, grant proposals, and guidelines. Each derivative work becomes another route by which the claim spreads.

Even after the notice appears, papers already accepted or in production may cite the old work, and a post-retraction citation date does not prove that the author knew the status when submitting, although a final editorial check could still catch it.

Researchers often cite without reading the source#

A plausible citation can be copied from a review, a colleague's draft, or another reference list. Citation errors that repeat across papers reveal this chain behavior. The new author may never open the original article, much less its publisher page.

Secondary citation also strips context. A review may summarize a result favorably, then the underlying study is retracted while the review remains unchanged. Later authors cite the review and unknowingly revive the invalid claim. Follow the claim back to the primary source, check that the source actually supports the sentence you are writing, and check its current status.

Old PDFs lose their warning labels#

Publisher HTML pages can display a banner, but a PDF saved before retraction may not update. Repositories and author websites may host an unmarked accepted manuscript. Screenshots and supplementary files can be detached from their landing pages.

Watermarking every accessible version helps, but publishers do not control every copy. A live DOI or PubMed check is safer than relying on the appearance of a local file. Reference managers can add alerts when their metadata connects to a retraction database. The warning still depends on correct identifiers and current synchronization.

Metadata does not propagate uniformly#

Publishers register articles and update notices in different systems. Crossref members can deposit relationships between a retracted work and its notice. Crossmark can show that an item has been updated. PubMed applies publication types and reciprocal links under National Library of Medicine policy.

Coverage is not identical. Some records lack a DOI or PMID, publishers may deposit updates late, titles can change, and databases ingest metadata on different schedules. Retraction Watch curates notices from publisher sites and is updated every working day, adding broad coverage where publisher metadata is incomplete. As of 2026, Crossref exposes those data through its production services and a downloadable dataset. Query more than one source when the answer matters.

A search result can hide the current status#

Search engines rank relevance, popularity, and textual match. An old PDF may appear above the retraction notice. Snippets can display the original abstract without the warning.

AI retrieval and summarization systems can inherit the same defect if their index stores article text but not update relationships; a confident synthesis is unsafe when source status is not a required metadata field. A system that gets this right resolves the persistent identifier at query time, surfaces corrections and retractions, records the date it checked, and stops a withdrawn result from silently supporting a claim.

Citation databases count intent poorly#

A paper may cite a retracted article to criticize it, describe misconduct, test citation decay, or explain why a field changed. Those are not instances of treating invalid data as valid.

Automated citation counts usually register a link, not the rhetorical function, and separating negative, historical, methodological, and affirmative use takes context analysis. So a headline such as "retracted paper cited 500 times" needs a denominator, dates, and citation context before you can read it as 500 errors.

Retraction reasons matter to reuse#

A paper can be retracted for fabricated data, major analytical error, plagiarism, redundant publication, unethical research, unauthorized data, compromised peer review, or another reason. Some notices concern the validity of findings directly. Others concern publication integrity while parts of the factual content may be available elsewhere.

The default is not to use a retracted article as evidentiary support. If a specific unaffected fact is historically necessary, the author should read the notice and any investigation carefully and explain the limited purpose. Retraction is not a universal accusation against every author. Notices should distinguish honest error, misconduct findings, and responsibility when known.

Systematic reviews face moving targets#

A review may screen thousands of records over a year, and a study can be valid when you search, retracted while you are extracting its data, and still sitting in your draft meta-analysis. Another can be retracted after you publish.

Review teams should store DOI, PMID, and other identifiers; check status at inclusion; repeat the check before synthesis and submission; and document the date and sources. Automated screening can flag records, but a human must read the notice and determine its effect. Living reviews need scheduled update checks. Static reviews may require a correction if removing a retracted study changes estimates or conclusions.

Retraction can propagate into guidelines#

Guidelines and decision aids often rely on systematic reviews. If a retracted trial contributed to a pooled estimate, risk model, or recommendation, removing one citation from the reference list is not enough.

The analytic effect should be recalculated. Recommendation panels need to ask whether certainty, balance of benefits and harms, or implementation advice changes. The update should state the affected evidence and new conclusion. One investigation found retracted trial reports embedded in reviews, guidelines, and later trials, with little downstream reassessment. Correction must follow the data, not stop at the original journal.

Citation checking belongs at several moments#

At literature search, use database retraction filters and status fields. During screening, resolve persistent identifiers and inspect publisher notices. During extraction, store status and check date with the study record.

Before you submit a manuscript, run your whole reference list through PubMed, Crossref, Retraction Watch data, or an institutional tool. Repeat the check after peer review, because months may have passed and the references may have changed. Editors and publishers can build the same check into submission systems. None of it is permanent, though: a paper you cleared today can be retracted tomorrow.

How to cite a retracted paper intentionally#

Sometimes the retracted work is the subject of analysis. State in the sentence that it was retracted, include the retraction year, cite the notice, and avoid language that treats the original finding as reliable.

The reference can include a bracketed retraction status according to the chosen citation style. NLM's Citing Medicine provides conventions for linking original items and notices. Do not hide the status in a footnote far from the claim. The reader should know before interpreting the sentence.

What authors should do when a cited source is retracted#

First work out how you used it. If it supplied background that other valid sources also support, replace it and check the wording still holds. If its data entered an analysis, rerun the analysis without it and look at how much moves.

If your conclusion changes, contact the journal promptly about a correction or other update. If it does not change, a transparent correction may still be needed so that your readers do not inherit the invalid citation. For a preprint or manuscript in review, revise directly and tell collaborators. Preserve an audit trail of the reason.

What journals and databases can improve#

Publishers should issue specific, freely accessible notices with persistent identifiers, bidirectional links, clear dates, and machine-readable update types. Every article format should display the status prominently.

Databases should ingest updates quickly and expose them through APIs. Reference managers, journal systems, evidence-synthesis platforms, and library resolvers can alert users at import and again when a library item changes. Notices should avoid vague language that leaves readers unable to tell whether findings are unreliable. Legal caution can coexist with proven facts and clear scope.

What readers can do in seconds#

Open the DOI on the publisher site and look for an update banner. Search the title or PMID in PubMed. Query Crossref or the Retraction Watch dataset if doubt remains. Read the actual notice, not only a third-party headline.

Check for a corrected and republished version, expression of concern, or later reinstatement. Status is more nuanced than a binary database flag.

Then ask whether the paper supports the claim for which it was cited. Valid status cannot repair a mismatched citation.