How to Use the Retraction Watch Database to Check a Paper or Journal

Most medical researchers know Retraction Watch exists. Fewer know they can search it directly, download the entire dataset for free, or have Zotero alert them the moment a paper in their library is retracted. The database has grown to more than 63,000 records as of early 2026, and since Crossref formally acquired it in September 2023 the infrastructure behind it has changed significantly. The search form, the bulk download route, and the REST API each work differently and suit different tasks. Understanding which to reach for, and when, is the practical question this guide is trying to answer.

There is a reason interest in this database keeps rising. In 2023 alone, more than 10,000 scholarly articles were retracted worldwide, a new record, and the number stayed elevated through 2024 and into 2025. Mass retraction events from publishers like Wiley, following the Hindawi acquisition, drove part of that count. The Retraction Watch database captured all of it. For researchers who spend months building a literature base for a systematic review, or who are about to cite a paper from a now-controversial group, the database is the most direct way to check what they are standing on.

What the Retraction Watch Database Is, and What Changed in 2023

Retraction Watch was founded in 2010 by science journalists Ivan Oransky and Adam Marcus as a blog tracking retractions and the stories behind them. The database grew out of that coverage: a structured log of retracted, corrected, and expression-of-concern papers, each manually checked and annotated by the Retraction Watch team before being entered. The records include the retracted paper's DOI, the journal, the author list, the retraction date, the type of notice, and one or more reason codes selected from a standardized taxonomy (data falsification, authorship disputes, plagiarism, error, and so on).

For years the full database required institutional access, which limited how many research teams could use it programmatically. That changed in September 2023 when Crossref acquired the database from the Center for Scientific Integrity and committed to maintaining it as an open, freely accessible public resource. Crossref took on the responsibility for hosting, updating, and distributing the data. The annotated records from Retraction Watch now flow into Crossref's infrastructure and are refreshed once per working day. That means the database you can download today contains the most recent retractions Crossref has processed, typically within one to two working days of a retraction notice being issued.

This matters for medical authors because Crossref is already the infrastructure that manages the DOIs of almost every peer-reviewed paper. The practical consequence is that retraction information and the paper's own DOI metadata now live in the same system. When you query a DOI through the Crossref API and that paper has been retracted, the API can return that information directly, without a separate lookup.

Using the Web Search Form

The simplest starting point is the web form at retractionwatch.com. The database search page lets you query by author name, journal title, DOI, PubMed ID, or a date range. You can use a single criterion or combine several. The form returns up to 50 entries per search, so for a journal or an author with a large retraction count you may need to narrow your query or use the bulk download instead.

In October 2024, Retraction Watch updated the rules for using the search form, requiring at least one specific criterion rather than allowing open-ended browsing. That change came because the database's growth (passing 55,000 records at the end of 2024) had slowed down broad, unfiltered queries. For practical purposes the requirement is not a significant constraint: a search by author surname, a specific journal, or a DOI will always return results quickly.

The form is most useful for targeted checks: looking up a specific co-author's publication history before agreeing to collaborate, verifying whether a paper you plan to cite has been retracted, or investigating a journal you are considering submitting to. For a journal-level check, type the journal name and limit the date range to the last five years. The results will show you not just the count but the retraction reasons, which can distinguish a journal that has handled data-integrity problems honestly from one where most notices say only "error" without further explanation.

How to Download the Full Dataset

For anyone who needs to check a long reference list, run a bulk audit across a body of literature, or keep a local copy of the database updated over time, the bulk download route is more practical than repeated web searches. Retraction Watch and Crossref maintain the full dataset as a CSV file in a public GitLab repository at gitlab.com/crossref/retraction-watch-data. The CSV is updated once per working day.

The simplest way to get the dataset is to clone the repository with git, which creates a local folder called retraction-watch-data and downloads the current CSV. Running a git pull from that folder on any subsequent day fetches only the changes since the last update, which is much faster than downloading the full file again. If you do not want to use git, the repository also offers a direct file download, though for a dataset of this size the clone-and-pull approach is easier to maintain over time.

The CSV contains one row per retraction or correction notice, with columns for the original article DOI, the retraction notice DOI (where one exists), the journal, the publisher, the article title, the author list, the retraction date, the notice type (retraction, correction, expression of concern, or reinstatement), and one or more reason codes from the Retraction Watch taxonomy. The reason-code column uses a semicolon-separated list when multiple reasons apply to a single notice, which is common: a paper retracted for data fabrication may also carry a code for authorship issues if the fraud investigation found evidence of both.

The Crossref REST API Route

For researchers comfortable making API requests, the Crossref REST API provides programmatic access to retraction data without downloading the full CSV. Retraction notices appear in Crossref's works records in the update-to field of the JSON response, and each record includes a source field that identifies whether a retraction entry came from the publisher directly or from the Retraction Watch team. That distinction matters: publisher-supplied retraction notices often have less annotation than Retraction Watch-curated ones, so the source field tells you how much accompanying metadata to expect.

A basic query filtering for retractions returns up to 100 records per request, with cursor-based pagination for larger result sets. The full Crossref documentation for retraction-related queries is at crossref.org/documentation/retrieve-metadata/retraction-watch. For a simple DOI-level check, you can query the works endpoint directly with a specific DOI and inspect the update-to field of the response to see whether any associated notices exist. This is the same mechanism that tools like Zotero, scite.ai, and several manuscript preparation platforms use under the hood.

If you are building a systematic review and want to screen a list of a few hundred papers, a short Python or R script looping through your DOI list and querying the Crossref API is more targeted than a full CSV download. The tradeoff is that the API returns data for papers with Crossref-registered DOIs, so papers published before widespread DOI adoption or in journals that do not register with Crossref will be missed. For recent biomedical literature, coverage is good. For older papers or grey literature, the web form or a manual check is more reliable.

Zotero's Built-In Retraction Watch Integration

If you use Zotero as your reference manager, you already have access to a version of this check that runs automatically. Zotero has integrated with Retraction Watch data since 2019, and after Crossref's acquisition it updated the integration in June 2023 to use the new Crossref-hosted dataset. The way it works is straightforward: Zotero stores a local copy of the Retraction Watch data on its servers and checks items in your library against it without uploading your library contents to any external service.

When a paper in your Zotero library matches a record in the Retraction Watch database, Zotero adds a warning flag to the item in the details pane, showing the retraction date, notice type, and a link to the Retraction Watch entry. If you are using the Zotero word processor plugin and you attempt to insert a citation for a flagged item, the plugin will stop and ask you to confirm the citation before inserting it. A paper that is retracted after you have already cited it in a draft will trigger the warning on the next time you update the citation.

The limitation is coverage. Zotero's integration works only for items with a DOI or a PubMed ID, which covers roughly three quarters of the Retraction Watch database. Papers without those identifiers will not be flagged. For most current biomedical literature this is not a significant gap, but older papers, conference papers, and some open-access preprints may not be checked. Mendeley does not have an equivalent built-in integration as of late 2026, so Zotero users have a meaningful advantage here.

What the Reason Codes Tell You (and What They Don't)

One of the most useful features of the Retraction Watch database, relative to a simple retraction count, is the reason taxonomy. Each retracted paper is tagged with one or more reason codes from a controlled vocabulary: categories include data fabrication, data falsification, manipulation of images, plagiarism, authorship problems, honest error, duplicate publication, concerns about peer review, and several others. A full list is available on the Retraction Watch website.

The reason codes matter because not all retractions have the same implications for the literature that cites the retracted paper. A paper retracted for a minor processing error that the authors themselves discovered and flagged is a very different situation from a paper retracted after an investigation found fabricated images or invented patient data. In the first case, the finding you cited may still be valid and simply needs a note in your manuscript acknowledging the correction. In the second case, any finding from that paper should be treated with caution even if the specific fact you cited was not the one found to be fraudulent.

There is a caveat worth naming. Retraction notices vary considerably in how much they disclose. Some journals provide detailed notices that explain exactly what was wrong and which data were affected. Others, particularly from publishers that have faced criticism for slow retraction processes, issue notices that are vague to the point of being uninformative. The reason code reflects what Retraction Watch could determine from publicly available information. When a notice says only "error" or "data concerns" without further detail, that reflects the opacity of the notice itself, not necessarily a judgment by the Retraction Watch team. In those cases, tracking down the original retraction notice, or looking for related post-publication commentary, gives a fuller picture.

Checking a Journal Before Submission

Retraction Watch can also be useful in the opposite direction: before you submit to a journal, checking its retraction history gives you a reading of how the editorial office handles integrity problems. A journal with a modest number of retractions for honest error and well-explained notices is behaving responsibly. A journal with many retractions for peer review manipulation, data fabrication, or paper mill activity, particularly over a short window, raises questions worth investigating before you invest months of revision work.

Search the database by journal name and restrict the date range to the past three years. Look at the retraction counts relative to the journal's publication volume (a journal publishing 5,000 papers per year and retracting 20 is in a different situation from one publishing 500 and retracting 20). Look at the reason codes: a cluster of notices for "concerns about peer review" or "paper mill" suggests systemic editorial problems, not isolated incidents. The Retraction Watch database alone will not tell you everything about a journal's quality, but it adds one concrete data point that most journal-selection advice ignores.

The same logic applies to checking a potential co-investigator or a research group whose work you plan to build on. A single retraction with a clear explanation is not automatically disqualifying. A pattern of retractions concentrated in one laboratory, or retractions for data fabrication where the retraction notice is unusually vague, deserves more scrutiny. The database gives you the record; the judgment is still yours.

Gaps to Know About

The Retraction Watch database is the most comprehensive resource of its kind, but it is not complete. Coverage depends on journals issuing formal retraction notices, and not all journals do so promptly or at all. Some corrections that effectively invalidate a paper's findings are published without the word "retraction" appearing anywhere in the notice, which means they may not appear in the database even though a reader of the correction would conclude the paper's main finding no longer stands. The database team does its best to capture expressions of concern and major corrections, but there is an inevitable lag between a problem becoming known and the formal notice appearing in the literature.

Coverage is also uneven across languages and regions. The Retraction Watch team primarily monitors the English-language literature and the major indexed databases. Retractions in journals indexed only in regional databases, or published in languages other than English, are likely underrepresented. For a systematic review that includes non-English sources or regional literature, the database is a useful first step but not a complete screen.

Finally, the database logs what was retracted, not what should have been. Papers under investigation, papers flagged on PubPeer with serious concerns, or papers where a post-publication reanalysis raised questions about the data do not appear in the database until a formal notice is issued. For a thorough integrity screen, combining a Retraction Watch check with a PubPeer search and a review of the citing literature (using tools like scite.ai, which records whether a paper was cited as supporting, contrasting, or mentioned) gives a more complete picture than any single tool alone.

Building the Check Into Your Workflow

The most reliable way to use the Retraction Watch database is to make the check routine rather than ad hoc. Ad hoc checks tend to happen only when something looks wrong, which is too late. A retraction check done at the point of submission, for every citation in the reference list, takes around thirty minutes for a typical clinical paper with fifty references, provided you have a working method already set up.

One practical structure: use Zotero with retraction alerts active as your primary reference manager throughout the project, which handles the ongoing monitoring. Before submission, export your final reference list as a CSV, extract the DOI column, and run a join against the locally cloned Retraction Watch CSV. Review any hits. For papers that return a result, read the retraction notice, confirm the notice type, and decide whether to remove the citation, add a note, or verify that the specific finding you cited was not implicated. Document your check in a brief note, because some journals may ask whether you verified your citations for retraction status as part of their submission checklist.

The retraction rate in the published literature is still low in absolute terms, which means that for most papers, this check will come back clean. But it costs very little, the infrastructure to do it is now free and publicly maintained, and the consequences of missing a retracted source in a clinical paper, particularly in a field where the retracted paper influenced a recommendation, are serious enough to make the check worth building in. The Retraction Watch database exists precisely to make this possible. Using it is straightforward. The harder part is making it a habit.

Further Reading