“Retraction” has a clean, one-line operational definition: the formal withdrawal of a published article by the journal, accompanied by a public notice explaining why. What that definition leaves out is everything that actually happens before a retraction notice exists — who decides, what has to be true for a paper to qualify, how an author gets notified, and how a retraction is different from the two outcomes editors reach for instead. This guide covers that process, using the current guidance from the Committee on Publication Ethics (COPE) and the International Committee of Medical Journal Editors (ICMJE), the two bodies most journals point to when a case actually arises. For how a retraction is registered and propagated once it’s decided — Crossref’s Crossmark service, the update-to metadata field, and whether it reaches a researcher’s ORCID record — see CASRAI’s companion guide on post-publication peer review, which covers that mechanical layer in detail. This page stays on the editorial decision itself.
What actually has to be true for a retraction to be warranted
Both COPE and ICMJE frame retraction as a remedy for a specific category of problem: the paper’s central findings or conclusions can no longer be relied on. ICMJE’s own guidance ties retraction to confirmed scientific misconduct — data fabrication, data falsification (including deceptive image manipulation), plagiarism, undisclosed major conflicts of interest that would have affected how the work was read, and methodology serious enough to invalidate the results. COPE’s retraction guidelines add further grounds beyond misconduct alone: redundant or duplicate publication without proper disclosure, use of data or material without authorization or in breach of copyright, and — critically — honest error, where the authors themselves discover a mistake serious enough that the paper’s conclusions are no longer trustworthy. Retraction is not a punishment; both bodies are explicit that its purpose is to correct the scholarly record, not to discipline the authors, which is why an author-initiated retraction for an honest, serious error is treated the same as an editor-initiated one for misconduct.
Just as important is what does not warrant retraction. If the paper’s main findings remain reliable and the problem is bounded — a wrong figure in one panel, a transposed digit, an author’s affiliation listed incorrectly — the appropriate outcome is a correction (corrigendum or erratum), not a retraction. COPE’s own framing of the decision is whether a corrected reader would draw materially different conclusions from the paper; if not, correcting the record in place is proportionate, and retracting a paper over a fixable, non-invalidating error is treated as an overreaction the guidelines specifically caution against.
Who actually makes the decision
ICMJE’s guidance states a preference for the original authors to retract their own work. In practice, that’s often not how it happens — authors may disagree with the finding, be unreachable, or (in misconduct cases) be the subject of the investigation itself. Where authors are unwilling or unable to retract, ICMJE guidance allows the editor to accept a retraction from another responsible party, or to publish the retraction under the editor’s own authority. Either way, final responsibility sits with the journal editor, not with COPE. COPE is a membership and standards organisation — it publishes guidelines, flowcharts, and case-by-case advice that editors can consult, but it does not itself investigate individual papers, adjudicate disputes, or order a journal to retract anything. The editor (in consultation with the publisher, and often the author’s institution where misconduct is alleged) is the actual decision-maker; COPE’s role is to supply the standard the decision should be measured against.
Institutions matter because they typically control evidence an editor can’t independently obtain — original data, lab records, interviews with the people involved — and a formal institutional research-misconduct finding often triggers or confirms a retraction. But the two processes are legally and organisationally separate: an institution can find misconduct without the journal having retracted anything yet (and vice versa), which is exactly why an expression of concern exists as an interim step — see below.
The process, step by step
The following sequence is a synthesis of COPE and ICMJE guidance, not a description of any single real case.
- A concern is raised. This can come from a reader, a co-author, a journal’s own pre- or post-publication screening, an institution, or an independent third party — including data sleuths and commentary platforms such as PubPeer. COPE’s current guidelines explicitly recommend crediting a third party who raised a valid concern in the eventual retraction notice, with their permission.
- The editor opens an internal review. This is not yet a determination — it’s an assessment of whether the concern is credible enough to warrant further investigation, and whether it can plausibly be resolved with a correction or requires treating as a potential retraction.
- Investigation, often involving the institution. For anything touching possible misconduct (as opposed to a straightforward author-reported error), COPE guidance directs editors to involve the author’s institution, since institutions typically hold the primary investigative authority and access to original records. Historically this step could stall a retraction for years while an institutional inquiry ran its course; COPE’s most recent guidance explicitly pushes back on open-ended delay, stating that timeliness should not be held hostage to an institutional investigation that isn’t moving or an uncooperative author.
- An interim expression of concern, if the matter is serious but unresolved. Where credible doubt exists but the investigation hasn’t concluded, the journal can publish an expression of concern — a separate, DOI-registered notice, linked to the original article, signalling that something is under review without yet reaching a retraction/no-retraction conclusion.
- A determination is reached and the retraction notice is drafted. Per ICMJE, the notice must be prominently labelled, carry the original article’s title in its own heading, explain the reason for retraction, and include a complete citation back to the original — and per COPE, it should reference any external investigation where relevant, avoid unnecessary detail that reads as punitive, and, where the retraction affects multiple related papers (for example, output tied to a single paper mill), state plainly that coordinated, systematic manipulation was identified rather than describing each paper in isolation.
- Authors are notified and, where possible, invited to respond or co-sign. Journals do not typically retract silently — authors are ordinarily given the opportunity to comment before publication, even where the editor ultimately proceeds without their agreement. In mass-retraction cases, COPE’s current guidance specifically recommends offering co-authors who were not implicated a chance at re-review before the notice is published.
- The notice is published and formally linked to the original. It gets its own DOI, is indexed so it’s discoverable independent of the original article, and — per COPE’s current guidance — should not sit behind a paywall even where the retracted article itself does. The original article stays in the public domain, clearly and permanently marked as retracted, rather than being removed.
Steps 5 through 7 are where COPE’s guidance and Crossref’s actual registration mechanism meet — see the post-publication peer review guide for how that notice then propagates via Crossmark’s update-to field, including the Retraction Watch Database integration Crossref has run since September 2023.
Retraction, expression of concern, and correction: not interchangeable outcomes
These three are often used loosely in casual conversation but describe three different editorial conclusions:
- Correction (corrigendum/erratum) — the paper’s conclusions are still sound; a bounded error is fixed in a linked notice. Conventionally “corrigendum” denotes an author-originated error and “erratum” a publisher-introduced one, though usage varies by journal.
- Expression of concern — an interim signal. The journal has a credible, unresolved concern but hasn’t yet determined whether a correction or a retraction is the right outcome. It’s meant to be temporary, resolving into one of the other two once the investigation concludes — though COPE’s own guidance acknowledges these can and do sit unresolved for extended periods where an institutional investigation stalls.
- Retraction — a conclusive determination that the paper’s findings can no longer be relied on, whether from confirmed misconduct or a serious honest error. The article is not deleted, but it is permanently and visibly marked as withdrawn.
The practical test COPE and ICMJE both point editors toward is the same one that separates correction from retraction above: would a reader who knew everything the editor now knows draw a materially different conclusion from the paper? A “no” points toward a correction; an unresolved “maybe” points toward an expression of concern; a clear “yes” points toward retraction.
What changed in COPE’s 2025 retraction guidelines
COPE replaced its 2019 retraction guidelines with a revised version in 2025, and the changes respond directly to problems that have become far more common since 2019 — most visibly, industrial-scale research fraud from paper mills. The current guidance explicitly lists paper-mill involvement, compromised peer review (fake reviewers or citation manipulation), unverifiable authorship, and undisclosed use of AI in the manuscript among the grounds for retraction — none of which the 2019 version addressed by name. It also adds process changes covered in the steps above: crediting third parties who identify problems, treating institutional-investigation delay as something editors should work around rather than wait on indefinitely, giving uninvolved co-authors a chance at re-review in mass-retraction cases, and requiring that retraction notices for coordinated manipulation say so plainly rather than treating each paper as an isolated incident. Anyone relying on older summaries of “the COPE retraction guidelines” that describe only the 2019 version should treat this update as the current baseline.
Frequently asked questions
Can an author refuse a retraction, or dispute one after the fact?
An author’s agreement is not required. ICMJE guidance prefers author-initiated retraction, but where authors are unwilling or unable to retract, the editor may proceed with the retraction under the journal’s own authority, or accept it from another responsible party. Authors are ordinarily given a chance to respond before a notice is published, and can raise a dispute with the journal or, where relevant, their institution — but the editor holds final responsibility for the journal’s content and the decision itself.
Does COPE retract papers, or just advise on them?
COPE does not retract papers, operate as a court, or overturn a journal’s decision. It is a membership and standards organisation that publishes guidelines and offers case-by-case advice to editors who bring it a specific situation. The decision to retract — and the responsibility for it — sits with the journal editor and publisher.
How is a retraction different from an expression of concern?
An expression of concern is provisional: a credible concern exists, but the investigation hasn’t concluded, so the journal isn’t yet stating whether the paper’s findings are unreliable. A retraction is conclusive: the journal has determined the findings can no longer be relied on. An expression of concern is meant to eventually resolve into either a correction or a retraction, not to stand as a permanent outcome on its own.
Is there a required deadline for issuing a retraction once a concern is confirmed?
No fixed timeframe is mandated in COPE’s or ICMJE’s published guidance — both acknowledge investigations vary too much in complexity to set one. COPE’s 2025 update does push toward faster action by discouraging open-ended delay tied to a stalled institutional investigation, but “as soon as the editor is confident the article is seriously flawed” is a standard, not a specific number of days or months.
Does a retracted paper get removed from the journal or from indexes like Crossref?
No. Both COPE and ICMJE guidance require the original article to remain accessible in the public domain, clearly and permanently labelled as retracted, rather than being deleted. It stays discoverable through its original DOI, with the retraction notice linked to it — the mechanics of that linkage (Crossref’s Crossmark service, the Retraction Watch Database integration) are covered in the post-publication peer review guide.
