The PDF in your reference library does not update itself. A retraction check has to look at live records, cover every reference and be repeated before proofs.
Why the copy in your library is not enough
A paper's status can change years after you first read it. The PDF in your folder and the record in your reference manager were captured on the day you saved them, so they keep showing the paper as it was. The live record is what matters.
Learning how to check if a paper is retracted is not optional diligence in medical publishing. The ICMJE Recommendations state that "authors are responsible for checking that none of the references cite retracted articles except in the context of referring to the retraction." For articles in MEDLINE-indexed journals, ICMJE names PubMed as the authoritative source for retraction information.
For senior authors this responsibility is concrete. You sign off on the whole reference list, including citations added by students or co-authors you did not check line by line. A retracted source found by a reviewer costs a revision round. One found after publication may need a correction. Our research integrity guide sets this check in the wider context of an author's obligations.
What counts as a retraction, and what does not
Databases and publishers use several labels for post-publication changes. Each one calls for a different response, so identify which you are dealing with before you act.
| Label | What it means | What it means for your citation |
|---|---|---|
| Retraction | The findings or conclusions should not be relied upon | Do not cite as evidence |
| Partial retraction | Part of the article, such as one figure or table, is withdrawn | NLM treats these as errata; check whether the withdrawn part is the part you cite |
| Correction (erratum, corrigendum) | An error was fixed; the article stands | Cite it, and check that the corrected version still supports your point |
| Expression of concern | An editor has flagged doubts, often pending investigation | Use judgement; consider disclosing the status |
| Withdrawal | Some publishers use this term for removing an article, often at an early online stage | Read the notice; treat it like a retraction unless it says otherwise |
| Retracted and republished | The article was retracted and a corrected version published | Cite the republished version |
| Reinstatement | A previous retraction has been reversed | Read the notice before relying on the article |
Two details catch experienced authors. First, according to the NLM policy on errata and retractions, PubMed does not distinguish between retractions for honest error and those for misconduct. The label alone tells you nothing about the reason. Second, the same notice can be classified differently by different sources. One database may record a notice as a retraction while another records it as a correction, so read the notice yourself rather than relying on any one label.
The sources to check, and what each one misses
No single source records every retraction. Each of the six below has a different blind spot.
PubMed
In PubMed, a retracted article carries the publication type Retracted Publication, and its record includes a "Retraction in:" link to the notice. The notice itself is labelled Retraction Notice and links back with "Retraction of:". Expressions of concern are linked the same way, with "Expression of concern in:" on the original article.
To search, use Retracted Publication[pt]. Use Retraction Notice[pt] for the notices. Expression of Concern[pt] finds the concern notices themselves, not the articles they concern, so it will not flag a cited paper. Older guides sometimes give the tag as "Retraction of Publication"; that search returns no results in PubMed today.
What it misses: anything outside PubMed, and anything the journal has not yet published as a citable notice. NLM creates these links based on notices the journal publishes, so a retraction announced elsewhere may not appear yet.
The Retraction Watch database
Retraction Watch collects retractions from publisher websites, database searches, web searches and community reports. Crossref acquired the database in September 2023 and made it publicly available. You can search it on the Retraction Watch Database website, but the site itself recommends downloading the data for serious use. The full CSV is published in a Crossref repository and is updated every working day.
It has two strengths. It covers disciplines and publishers outside PubMed, and each record includes the reasons for the retraction using a controlled vocabulary.
What it misses: Crossref's documentation notes that expressions of concern and corrections are included but are not as comprehensive as retractions. Do not treat an absence of concerns in this dataset as proof that none exist.
Crossref metadata
Since January 2025, Retraction Watch retractions have also been available through the Crossref REST API. Look up a single DOI at https://api.crossref.org/works/ followed by the DOI. In the original article's record, an updated-by entry with "type": "retraction" indicates a retraction. Its source field shows whether the information came from the publisher or from Retraction Watch. The Crossref Retraction Watch documentation explains both access routes.
What it misses: works without a Crossref DOI, which includes much older literature and some regional journals.
Crossmark on the publisher's page
Crossmark is the Crossref button that publishers place on article pages and PDFs. Clicking it shows the current status of the article, including any corrections, retractions or other updates the publisher has registered. Because it can be embedded in a PDF, it can alert you to a change after you downloaded the file, provided you are online when you click it.
What it misses: Crossmark is optional. It only reports updates that the publisher has registered, so a page without a Crossmark button, or with one showing no updates, is not evidence that the paper is sound.
The publisher's article page
ICMJE expects retracted articles to remain available and to be clearly labelled as retracted in all forms, including the abstract, full text and PDF. Opening the live article page is therefore a quick confirmation, and it is the only check available for papers that have no DOI or PMID.
What it misses: practice varies between publishers, and some older or smaller platforms label retractions inconsistently.
Your reference manager
Zotero flags retracted items using Retraction Watch data. It warns you if you try to cite one through its word processor plugin, and again when you refresh citations that are already in a document. Recent versions of EndNote offer a similar retraction alert when you are signed in to an EndNote account.
What it misses: both tools match on DOI or PMID, so references without those identifiers are not checked. Treat these alerts as an early warning, not as the final check.
A batch check for a whole reference list
Checking 60 references one at a time is how the check gets skipped. This six-step batch method takes far less time.
- Collect identifiers. Export the DOI and PMID for every reference. List the references that have neither separately, because the automated checks will not cover them.
- Run a batch PubMed search. Join the PMIDs with OR and add the filter, for example:
(12345678[pmid] OR 23456789[pmid]) AND Retracted Publication[pt]. Any result is a hit. This search does not catch expressions of concern. Those appear only as an "Expression of concern in:" link on each article's PubMed record, so open the records or rely on theupdated-byentries in step 3. - Check DOIs against Retraction Watch data. Match your DOI list against the OriginalPaperDOI column of the downloaded CSV, or look each DOI up in the Crossref API and look for
updated-byentries. - Check the references without identifiers. Search their titles in the Retraction Watch database, then open each one on the publisher's site and look for a notice.
- Read every notice you find. Note the scope (full or partial), the stated reason, and whether a corrected or republished version exists.
- Decide, record and repeat. Revise the manuscript for each hit. Keep a dated log of what you searched and found. Run the check again before you approve proofs, because status can change while a paper is under review.
If the manuscript has a long gap between submission and acceptance, run the check again on the references that were added during revision. These are the ones most likely to have skipped the first check.
What to do when you find a hit
A retracted source used as evidence
Remove it as support for the claim. Do not simply swap in another citation. First ask whether the claim still stands without the retracted paper. If the retracted study was the main evidence for a statement in your introduction or discussion, the statement may need to be softened, qualified or removed.
A retracted source you need to discuss
Citing retracted work is acceptable when the retraction is the point, for example in a paper about the history of a claim, the influence of flawed studies, or research integrity itself. Say in the text that the work was retracted, and cite the retraction notice as well as the original. PubMed records add "Retraction in:" followed by the notice details. Follow your target journal's reference style for the exact format, because styles differ.
An expression of concern
The outcome is not yet settled. Consider whether you need the source at all. If you do, disclose the status in the text so readers and reviewers can judge for themselves.
A correction
Check whether the correction changes the number, figure or conclusion you cite. If it does, update your text to match the corrected version.
Evidence syntheses
In a systematic review or meta-analysis, a retracted included study affects the results, not just the reference list. Re-check your included studies before submission. If one has been retracted, state how you handled it and consider reporting the analysis with and without it.
When a source is retracted after your paper is published
Retractions sometimes come years after the original paper, so a source can be retracted after your own article is published. If your conclusions do not depend on it, you may not need to do anything. If they do, contact the journal that published your article and explain the situation. The editor can decide whether a correction is warranted. Our guide to correcting the published record covers how that process works.
Keep in mind that the Committee on Publication Ethics describes the purpose of retraction as correcting the literature, not punishing authors. The same applies to your response. Updating your own record promptly is good practice, not an admission of misconduct.
Making it routine in your group
A check that depends on one careful person will eventually be missed. For senior authors and lab leads:
- Assign the retraction check to a named co-author for every submission
- Require DOIs or PMIDs in the shared reference library wherever they exist
- Turn on retraction alerts in the group's reference manager
- Keep the dated check log with the submission files
- Re-run the check at revision and before proofs
- Teach students to read the notice, not just the label
Building this into lab routine is part of the wider set of research integrity responsibilities of a lab lead. It takes a few minutes per manuscript and protects every name on the author list.
When your reference list is checked and your manuscript is ready, the Directive Publications author guidelines explain what to prepare before you submit.
Frequently asked questions
How do I check if a paper is retracted?
Search the paper by DOI or PMID in PubMed and look for a Retracted Publication label or a Retraction in link, then check the DOI against the Retraction Watch data that Crossref makes openly available. Finish by opening the article on the publisher's site and looking for a retraction notice or Crossmark update. No single source is complete, so use more than one.
Can I cite a retracted paper?
You can cite it when the retraction itself is relevant, for example when you discuss the retracted work, its influence or the reason it was withdrawn. You should not cite it as supporting evidence for a claim. When you do cite it, say in the text that it was retracted and cite the retraction notice in the style the journal requires.
Is the Retraction Watch database free to use?
Yes. Crossref acquired the database in September 2023 and made it publicly available. You can download the full dataset as a CSV file, which is updated every working day, or look up individual DOIs through the Crossref REST API.
What is the difference between a retraction and an expression of concern?
A retraction tells readers that a paper's findings or conclusions should not be relied upon. An expression of concern is an editor's warning that questions have been raised, usually while an investigation is under way, and it may later be followed by a retraction, a correction or no further action. Treat an expression of concern as material context that you may need to disclose.
Does every journal article have a Crossmark button that shows retractions?
No. Crossmark is an optional Crossref service, and it only shows an update when the publisher participates and registers that update. A missing Crossmark button tells you nothing about the paper's status, so it should never be your only check.