Editorial Workflows & Citations

Is This Citation Still Current? A Prepublication Status Check

A DOI can still resolve after the evidence behind it has changed. Here is a repeatable way to catch corrections, expressions of concern, and retractions before a source-heavy document goes out the door.

September 4, 2026 10 minute read

Citations age even when the prose around them does not. A paper that was current when a writer saved the PDF can acquire a correction six months later, an expression of concern during review, or a retraction after a course, manuscript, or clinical document has entered production. The reference may remain perfectly formatted. The link may still work. Neither fact tells an editor whether the cited evidence is still safe to use.

That creates a quiet gap in many editorial workflows. Teams verify a source when they capture it, then spend weeks checking language, numbering, style, and layout without checking the source’s publication status again. A final status check does not replace scientific or subject-matter review. It does catch a different class of problem: the scholarly record changed while the document was moving toward release.

The practical takeaway

Check citation status twice: when evidence enters the project and again shortly before publication. Use more than one signal for high-stakes claims, record what was checked, and route every flag to a human decision.

Why this check is more practical now

In January 2025, Crossref moved Retraction Watch records into its production REST API, combining curated Retraction Watch data with retraction and correction metadata supplied by Crossref members. In May 2026, Crossref formally marked the earlier Labs experiment as retired and warned that the Labs endpoint was no longer current. The production service now gives editorial systems one open, machine-readable place to look for many post-publication updates. Retraction Watch continues to update its underlying dataset every working day.

The infrastructure is better, but the risk is not theoretical or limited to deliberately deceptive research. A 2025 systematic review in JAMA Internal Medicine identified 61 systematic reviews that had included at least one retracted study. In 37 of them, the retraction happened only after the review was published. Removing the retracted study changed statistical significance in 18 of 166 recalculated meta-analyses. The lesson for an editorial team is simple: a source can change status after a careful evidence review, and sometimes that change matters to the conclusion.

A status flag is not a binary verdict

Before building a check, distinguish the signals it can return. The 2024 NISO recommended practice for communicating retractions, removals, and expressions of concern focuses on making those statuses clear to both people and machines. Its categories matter because they call for different editorial responses.

A correction or erratum says that published content changed. The correction may be minor, or it may alter a number, method, attribution, or conclusion your sentence depends on.
An expression of concern is a warning that serious questions remain unresolved. It is not the same as a retraction, but it raises the threshold for relying on the work.
A retraction says the publication should no longer be treated as part of the reliable literature for its original claims. A retracted paper may still be cited when the retraction itself is the subject.
A removal or withdrawal has its own context. The notice and publisher record should explain what happened; an absent PDF alone is not enough information.
A new version or republication may supersede the copy in the project folder. The International Committee of Medical Journal Editors recommends citing the most recent version.

Do not design the workflow so that “any update” automatically deletes a reference. Design it so that the update, affected claim, replacement options, and accountable reviewer meet in one place.

A six-step prepublication citation-status check

Step 1

Normalize the reference and preserve its identifier

Start with the DOI when one exists. Remove resolver prefixes, trailing punctuation, and accidental spaces, but keep the original reference text for comparison. For biomedical sources, also retain the PMID. For standards, guidance, books, and web sources without a DOI, save the stable publisher or issuing-body URL, title, date, and version. Identifier cleanup is not clerical polish; it determines whether later checks find the correct record.

Step 2

Open the record of publication

Visit the DOI landing page or the publisher’s canonical article page, not only a saved PDF or copied abstract. Look for a Crossmark button, banner, linked notice, version date, or publisher note. Crossmark is designed to expose significant updates—including corrections and retractions—on the web and inside participating PDFs. Read the linked notice rather than relying on the color or label alone.

Step 3

Check the specialist index that serves the subject

For biomedical literature, inspect the PubMed record. The U.S. National Library of Medicine creates links between original citations and related notices, including errata, retractions, expressions of concern, and corrected-and-republished articles. Those two-way links often make the relationship clearer than a PDF stored months earlier. Other disciplines may have a trusted index or library discovery service with similar status signals.

Step 4

Query the production metadata

Look up the DOI through Crossref’s production REST API and inspect the update relationships. Crossref reports retractions in structured fields and identifies whether a signal came from a publisher or Retraction Watch. For batch workflows, use the production API or the current Retraction Watch dataset—not the retired Labs endpoint. Treat missing metadata as “no flag found,” not as proof that no update exists.

Step 5

Map the notice to the exact claim

Highlight every sentence, table cell, figure, learning objective, or recommendation supported by the flagged source. Then read the notice and corrected material with the subject-matter reviewer. A corrected author name may leave the claim untouched. A corrected denominator, outcome, or interpretation may require new prose, new evidence, or a fresh approval. The decision belongs at the claim level, not only in the bibliography.

Step 6

Record the check and close the loop

Keep the identifier, systems checked, check date, status found, notice URL, affected content locations, reviewer, decision, and replacement source if used. Re-run cross-references and numbering after changes. Then confirm that the corrected reference appears consistently in the Word file, slides, eLearning export, structured XML, and any derivative handout.

What to do when the check finds something

If the source has a correction

Update the citation to the current version, read the correction, and compare it with the claim. If the claim still stands, record why. If a number, population, method, or conclusion changed, revise the dependent content and send it back through the appropriate review. Do not silently swap the bibliography while leaving the prose untouched.

If the source has an expression of concern

Escalate it. For a peripheral background statement, a qualified reviewer may decide to replace the source or add context. For a clinical, safety, compliance, or central learning claim, pause release until the evidence owner decides whether independent support is sufficient. The expression is a reason for scrutiny, not permission to predict the investigation’s outcome.

If the source is retracted

Do not use it as affirmative support for its original findings. Identify every downstream reuse, including copied tables, speaker notes, alt text, facilitator guides, question feedback, and localized versions. If the document discusses the retraction, cite both the original item and the retraction notice so readers can understand the context. A replacement source must support the actual sentence; a similar title is not enough.

If no flag appears

Continue the ordinary evidence review. Crossref notes that Retraction Watch coverage is strongest for retractions and less comprehensive for some other update types. PubMed covers biomedical literature, not every publication. Publishers differ in how quickly and consistently they expose notices. No single database is a universal seal of validity, which is why high-stakes work should combine the publisher record, relevant index, and structured metadata.

Make it a system, not a heroic final search

The best time to design this control is before the reference list reaches hundreds of entries. Require stable identifiers at evidence intake. Store each source once and link its internal record to every place it is used. Give writers a clear status vocabulary and an escalation owner. Then place checks at three useful gates:

Intake: confirm the identity and current status before the source enters the approved evidence set.
Prepublication: recheck all cited sources—or at minimum every source behind a consequential claim—as close to release as the workflow permits.
Maintenance: recheck evergreen courses, guidance, and high-use publications on a defined schedule or when an update alert arrives.

Automation can retrieve identifiers, compare metadata, and flag changes. It should not decide whether a correction changes the meaning of a clinical statement or whether an expression of concern leaves enough independent support. That judgment needs editorial context and qualified subject-matter review. The useful division of labor is simple: machines make the changing record harder to miss; people decide what the change means.

For citation-heavy teams

Keep the reference layer ready for review.

Superscriptify helps teams standardize and repair citation-heavy Word and PowerPoint files, align references for JATS workflows, and carry cleaner source structure into editorial review.

Explore JATS reference alignment

Sources and further reading

This article provides general editorial and educational information, not medical, legal, or regulatory advice. Decisions about clinical, safety, legal, or regulated claims should involve appropriately qualified reviewers.