Is the paper I cited retracted? How to check
Paste the DOI or the exact title into the Retraction Watch Database. Free, no account, about ten seconds. If the paper is there, read the notice before you do anything else, because what it says changes what you should do next. Be clear about what a clean result buys you, though. This is the cheapest of all the citation checks and, on a normal undergraduate reading list, the one least likely to find anything at all.
The rest of this page is the practical procedure: where else to look when the database comes back empty, how to tell a retraction from the two things people confuse it with, and what to do when a source you have already cited turns out to have been pulled.
The four-place check
One place is usually enough. Four makes it as certain as this can get, and each catches something the others miss.
The Retraction Watch Database
Search by DOI or by title at retractiondatabase.org. The search page has two boxes: use "Original Paper" for the article you are citing, and "Retraction or Other Notices" if you are looking for a notice you already know about. The user guide covers the other fields, including author, journal and PMID. You have to fill in at least one field, so a completely blank search will not work.
Crossref acquired the database from the Center for Scientific Integrity in September 2023 and made the data publicly available, refreshed once per working day and downloadable in full from a public repository. That matters for two reasons. It is why reference managers can now check retraction status automatically, which is the last section of this page. And it means that if you ever need to check a long list at once, a few hundred references against one CSV, you can, without asking anyone's permission or paying for access.
The article's own page
Resolve the DOI and look at the publisher's landing page for the article. A retracted paper normally carries a banner across the top, a "RETRACTED" prefix on the title, or a linked notice. Many publishers also show a Crossmark button, which reports whether Crossref holds a record of an update, correction or retraction for that DOI. This is the check that catches very recent retractions before third-party databases pick them up.
You can read the same thing without the button. Open https://api.crossref.org/works/ plus the DOI and look for an update-to field: that is where Crossref records that this article has been retracted, corrected or otherwise updated. One detail there repays a second look. Each entry carries a source, which is either publisher or retraction-watch, so the metadata tells you not just that the paper was retracted but who put that fact into the record. A retraction present as retraction-watch and absent as publisher is one the publisher has not surfaced in its own metadata, which is exactly the case where the article's own page can look perfectly clean. The whole flagged set is one query away too, with https://api.crossref.org/works?filter=update-type:retraction.
PubMed, for medical and life sciences
Find the record in PubMed and look at its publication types. Two labels matter: "Retracted Publication" is applied to the paper itself, and "Retraction Notice" marks the notice. Watch that second one, because NLM renamed it in MeSH 2025 and most guides still print the old name: searching "Retraction of Publication"[pt] today returns zero results, which reads exactly like a clean paper. The live counts, checked 11 August 2026, are 33,720 records tagged "Retracted Publication" and 32,764 tagged "Retraction Notice". You can also filter a whole result set by article type. The MD Anderson research library has a short walkthrough if you want the exact clicks.
OpenAlex, for a one-URL answer
OpenAlex carries a boolean is_retracted on every work, which makes this the fastest check of the four when you have a DOI: open https://api.openalex.org/works/doi: plus the DOI and read the field. It is also filterable, so https://api.openalex.org/works?filter=is_retracted:true returns the whole flagged set, which is how we got the number in the next section. Since 13 February 2026 OpenAlex asks for a free API key for more than light use.
Why your PDF said nothing
Because the file on your laptop is a snapshot and the notice attaches to the version of record. A copy downloaded eight months ago will never tell you anything changed. Check the live page, not your download.
Where the retraction record is incomplete
Every guide to this, ours included until now, tells you to search a database and stops. The databases are good and they are not complete, and the three gaps are independent of each other, which is why checking a second place is not redundant.
A retraction only exists because somebody issued one. This is the gap that swallows the others. A retraction database is a record of decisions journals took, not a record of papers that turned out to be wrong. A paper whose problems were never raised, or were raised and never investigated, or were investigated by a journal that decided to do nothing, is absent from every database on this page and indistinguishable from a clean one. No amount of cross-checking fixes that, and it is a hard floor on what this check can ever tell you.
The notice can exist while the record does not say so. A retraction is issued by the journal and then has to propagate into publisher pages, indexes and reference managers, and it does not always arrive. We measured this ourselves on 11 August 2026. Taking 300 retractions at random from the Retraction Watch dataset (2015 to 2025, original DOI present) and asking Crossref about each one, 96% were flagged as retracted. But only 73.3% were flagged by the publisher. The remaining 22.7% show up only because Retraction Watch fed them in, and 1% resolved with no retraction flag from anyone. So a quarter of what you can currently find out is there because one small organisation did the work, and the answer you get from a publisher alone is wrong about one time in four. Silence in one place is not a clearance, which is the whole argument for looking in more than one.
The notices themselves are patchier than the flags. Across the full dataset, 18.9% of retractions carry a notice problem: limited or no information about why, no notice at all, or a notice locked behind the publisher's paywall. And 8.9% have no DOI for the original paper, so they cannot be found by the check this page opens with. Neither figure appears in a published paper; both come from counting the dataset's own columns, and you can reproduce them from the same file.
Your own copy never updates. The version on your disk, in your reference manager, or in a course pack is frozen at the moment you saved it. This is the gap that catches the most careful students, because they are the ones with a well-organised folder of PDFs from eight months ago.
Now the numbers that should shape how you use all of this. On 11 August 2026, OpenAlex held 133,790 works flagged as retracted out of 323,998,640 in total. That is roughly four in every ten thousand.
The figure people quote instead is the record year. Nature reported that more than 10,000 papers were retracted in 2023, smashing the previous annual record, and it is usually cited as evidence that the literature is falling apart. Read the composition before you accept that reading. Most of that year's total came from a single source: Hindawi, then a Wiley subsidiary, retracted over 8,000 articles after its peer-review process was compromised. A record driven by one publisher cleaning up one breach is a different fact from a record driven by a general collapse, and it points the opposite way about the system's health. It also tells you something practical, which is that retractions cluster hard by publisher, by journal and by author group, so the useful question is rarely "has this paper been retracted" and usually "is this paper standing next to papers that have been".
So, an honest opinion about where this check belongs, and it is not where the enthusiasm usually puts it. Retraction checking is the cheapest of all the citation checks and, on a typical undergraduate reading list, the least likely to find anything at all. Run it, because ten seconds is ten seconds. Do not run it first, and never let it stand in for the checks that will actually catch something: does the paper exist, and does it say what you cited it for. It genuinely earns its place in three situations. An older paper your whole argument rests on. A finding that looks suspiciously clean for how hard the question is. And a field you know has a mill problem. The failure we would most like to prevent is the student who ran a retraction check on all fifteen references, found nothing, felt reassured, and never opened one of the papers.
Retraction, expression of concern, correction
Three different notices, routinely mixed up, and only one of them means stop citing this.
Correction (also erratum or corrigendum). A specific error is fixed and the findings still stand. Cite the paper normally. If the correction touches the exact number or figure you are relying on, use the corrected value.
Expression of concern. The editors are telling readers about a credible problem that has not been resolved, often while an investigation runs. It is not a retraction and it is not a clearance. Treat the finding as unsettled: either avoid resting an argument on it, or cite it and note the concern. Expressions of concern are frequently superseded later by either a correction or a retraction, so check the date and look for anything more recent.
Retraction. The work should no longer be relied on. The article stays online, marked, because the record is not erased. COPE's retraction guidelines set out when journals should retract and what a notice must contain. The 2025 update added categories including retraction with replacement, for serious errors that change the findings without invalidating the underlying scholarship, and retraction with removal for rare legal or public-health cases.
Reading the notice
Open it. It is usually one paragraph and it tells you four things worth knowing.
- Who issued it. The authors, the editors, the publisher, or some combination. An author-initiated retraction after finding their own error is a different story from an editorial retraction following an investigation.
- The grounds. Honest error, unreliable data, image manipulation, plagiarism, unverifiable authorship, ethics approval problems. COPE's guidance covers retraction for honest mistakes as well as for misconduct, so a retraction is not automatically an accusation.
- The scope. Some notices retract a specific finding or figure while noting the rest holds. Read carefully before you assume everything in the paper is void.
- The date. This determines whether the paper was already retracted when you cited it, which is the fact that shapes the conversation if anyone asks.
Some notices are vague to the point of uselessness. That is a known problem in the literature and not something you can fix. When the notice says nothing, treat the paper as unusable for evidence and move on.
What to do about it
Before you submit. Remove it and re-support the claim from a source that stands. Do not just swap in the nearest paper on the topic: check that the replacement actually says what your sentence claims, which is a separate test from whether it is sound. If nothing supports the claim any more, that is worth knowing, and the search-and-rewrite procedure covers what to do next.
After you submit, before it is marked. Email your instructor. One short message: the reference has been retracted, here is the notice, here is the corrected version if you are allowed to provide one. Getting there first is always better than being told.
If the retraction post-dates your submission. This is a normal hazard of writing about live research and no one reasonable will hold it against you. Say when you cited it and when the notice appeared. The dates do the work.
If you are writing about the retraction itself. Cite it explicitly as retracted, with the notice alongside. Discussing a withdrawn paper as a case is legitimate; presenting its findings as current evidence is not.
One extra step worth taking: check the rest of the list. Papers cluster, and a retracted source is often cited alongside others from the same group or the same problematic literature. While you are there, run the cheaper checks over the whole bibliography too. Paste it into our free citation checker, which looks up every entry in CrossRef and OpenAlex and reports separately whether the identifier resolves and whether the record behind it matches what you wrote, or push a single suspect identifier through the DOI lookup to see the exact record it is registered to. Neither needs an account, and between them they catch the two failures that are far more common than a retraction.
Making it automatic
If you use Zotero, retraction checking is built in. It checks your library against Retraction Watch data, flags retracted items in the item list, and warns you when you next update citations in a document that cites one. Setting that up once removes this task from your life permanently. Other reference managers have added similar features, and the underlying data is openly published by Crossref if you want to check a long list programmatically.
Where this fits into checking a reference at all: retraction status is the last of several questions, after whether the paper exists and whether the details are right. Our guide to verifying a citation is the full routine, and how to tell if a journal article is fake covers a related question this check does not answer, which is whether the venue was legitimate in the first place. A retraction database only knows about papers a journal cared enough to withdraw, which is one more reason to prefer sources from venues that behave like this in the first place.