RESEARCH EVIDENCE · CHECKING A CITATION

How to check whether a research paper has been retracted or corrected

A company brochure points to a scientific paper. The link works, the journal looks familiar, and the PDF still contains the impressive number. There is one more place to look: what happened to that paper after publication?

Start with the exact paper on the publisher’s website, inspect its current status and linked notices, and compare any change with the particular claim you plan to use. A correction may leave that claim intact or change it completely. A working DOI link and a familiar title settle neither question.

Keep the claim beside the citation

Before searching, copy the sentence you are checking and note where it appears. “This material lasted 48 hours in a salt-spray test” is a manageable claim. “This company is scientifically credible” bundles too many questions into one investigation.

Build a small identification card: paper title, authors, journal, publication year, DOI if present, and the version you actually read. Add the page, table or figure behind the claim. If the brochure gives only a screenshot, search the exact title and match several details. A similar title from the same research group can still be a different study.

Keep a preprint, an accepted manuscript and the published journal article separate in your notes. Crossref’s version guidance distinguishes these stages and explains how significant later notices connect to the work they update. Follow those connections rather than treating every PDF as the same object.

Look in three places, in this order

1. The publisher’s article page. Follow the paper’s DOI or the journal’s own search. Read status banners, linked corrections and editorial notices. Open the notice itself; a search-result snippet can omit the sentence that explains the change.
2. Crossmark, when available. Click the Crossmark button on the article page or PDF. It exposes publisher-supplied status and update information. Its presence is not a quality guarantee, and its absence is not evidence that a paper is invalid. Crossref explains what the button does.
3. A separate retraction lookup. Use the public Retraction Watch data described by Crossref, matching the original paper’s DOI or identifying details. Use any result to find the underlying notice. Crossref says coverage of corrections and expressions of concern is less comprehensive than coverage of retractions, so a missing result cannot clear every kind of update. See access methods and coverage.

In the CSV download, match OriginalPaperDOI, then follow RetractionDOI or the linked notice. If its DOI is missing, match the notice’s title and authors.

A useful additional web search is the exact title in quotation marks plus correction, retraction or expression of concern. This helps discover a notice that was hard to find on the landing page. It does not replace matching that notice to the right article.

Read the notice as an explanation, not a red or green badge

Do not turn the word “retracted” into an accusation against a named person. The 2009 COPE guidance on retractions explains that unreliable content can arise from honest errors as well as misconduct. Report the reason the notice gives. An unresolved concern also deserves different wording from a completed finding.

A worked example: the number survives online, but not the correction

Fictional training example. A made-up supplier, Alder Materials, advertises that a coating remained intact for 48 hours in a laboratory test. Its brochure links to Paper A. This is an invented exercise, not a claim about a real business, product or study.

You locate two notices. The first corrects an author’s institutional address. The second changes Table 2: the relevant test duration was 36 hours, not 48. The brochure still quotes 48, and an old downloaded PDF still displays it.

The address correction does not, by itself, alter the advertised duration. The table correction does. Your finding can therefore be precise: “The brochure’s 48-hour figure does not match the corrected 36-hour result in Table 2.” You have found a citation discrepancy. You have not established that the supplier deliberately lied, that every claim in the brochure is false, or that the product performs badly in every setting.

Now add one twist. Suppose the corrected study tested a different coating formulation from the one being sold. Even the corrected 36-hour figure would need a separate applicability check. Updating a number and establishing that a study supports a commercial claim are different pieces of work.

Write a claim-to-notice impact card

This short structure makes your decision reviewable by someone who has not followed every tab:

Claim being checked:
Where the claim appears, and capture date:
Exact paper and version:
Supporting table, figure or passage:
Status sources checked, and check date:
Notice title, URL/DOI and date:
What the notice changes:
Does the claim rely on that element? Why?
Decision: usable as stated / revise / withhold / unresolved
Remaining question:

For the fictional example, the decision is “revise or withhold the 48-hour claim,” with the corrected table linked. Preserve the old citation as part of the history instead of silently replacing it in your research notes. That lets another reader understand why your conclusion changed.

If sources disagree, record the disagreement. A publisher notice may be newer than a database entry; a local PDF may predate both. Compare the notice date, the copy’s version and the time you checked. When a notice cannot be read, do not guess its reason from an alarming headline.

Know what this check has and has not answered

Finding no retraction does not validate a study’s design, statistics or relevance. Finding a correction does not automatically make the whole paper unusable. This workflow answers a narrower and useful question: what is the current publication record, and how does it affect this specific citation?

If several pages repeat the same old claim, count the underlying evidence rather than the number of websites carrying it. Our source-verification exercise shows how to separate independent support from repetition.

For a single paper, the public tools above may be enough. If a company’s product pages, archived claims and cited studies contradict one another, prepare the exact claim, source links and desired decision before asking for a scoped OSINT Jet manual investigation. Ask whether that evidence question can be covered; this is not an offer of automatic scholarly peer review. A useful brief should make the unresolved connection visible, not just ask whether the company is “trustworthy.”