Reference and Citation Check for Manuscripts

Start with a table that has one row for each in-text citation. In each row, put your sentence, the claim in a few words and the reference it points to. Leave columns for the passage you find, its page, a verdict and the fix.

Next, reconcile the text with the reference list. Flag every citation that has no entry in the list, and every entry the text never cites. Both creep in when co-authors add and cut references during revision.

If you drafted from an evidence matrix, such as the one a literature review synthesis builds, the matrix may already hold the page behind each claim.

Before you judge a claim, confirm you have the paper you meant to cite. The International Committee of Medical Journal Editors (ICMJE) says references should be verified using either a bibliographic source, such as PubMed, or original sources¹. Match the authors, year, title and journal, and watch for the classic slip: the right author, the wrong paper.

Fix each error where the reference is stored, such as your reference manager or annotated bibliography, not just in the draft. Otherwise the same wrong year comes back in your next paper.

Take extra care with any reference a chatbot suggested. In a 2023 study in Scientific Reports, 18% of the references GPT-4 produced for short literature reviews did not exist². Most of the invented articles were credited to real journals, so a familiar journal name proves nothing. If you can't find a reference in the journal or a library database, don't cite it.

This is the check that matters most, and the one formatting tools skip. The ICMJE says authors should be able to attest that the references cited support the associated statement¹. Open the full text rather than the abstract, find the passage and judge your sentence against it.

Expect to find problems. A 2017 review in PLOS ONE pooled 15 studies of quotation accuracy in medical research articles and found errors in 14.5% of the cited claims those studies examined³. About two thirds were major: the source didn't support the claim, was unrelated to it or contradicted it. The rest were oversimplifications, overgeneralizations or trivial slips.

Sometimes the passage you find is your source citing someone else. A review's summary of a trial is not the trial, and it may have dropped a condition along the way. The ICMJE asks authors to provide direct references to original research sources whenever possible¹, and warns that review articles do not always reflect original work accurately.

Copy a citation from another paper's reference list and you copy its errors too. In a 2003 paper, two UCLA researchers modeled how the same misprints spread through citations of one famous physics paper. They estimated that only about 20% of citers read the original⁴.

So follow each chain to its end. Open the paper your source cites, find the passage and judge it like any other citation. If you can't get the original, cite the paper you read and make clear it reports someone else's finding.

Papers change after you cite them. The ICMJE makes authors responsible for checking that none of the references cite retracted articles¹, except when the retraction is the point. For journals indexed in MEDLINE, it treats PubMed as the authoritative source on retractions and suggests searching it for "Retracted publication [pt]".

For journals outside MEDLINE, search the Retraction Watch Database. Its user guide calls it the most comprehensive source on retractions, but says it does not aim for comprehensiveness in expressions of concern or corrections⁵.

That gap matters. A correction leaves a paper in print but changes something in it, and the change may be the number you quoted. Look for correction notices on each article's page at the publisher.

Preprints need one more step. Find out whether each one has been published, and compare every number you quoted with the journal version. A 2022 study in PLOS Biology compared early-pandemic preprints with their published versions. Most abstracts changed little, but 17.2% of COVID-19 abstracts and 7.2% of others had a major change in their conclusions⁶.

The tool at the top of this page does the first pass for you. It quotes the passage behind each citation beside your sentence and gives a verdict wherever it can read the source in full. It also checks every reference's status and leaves the edits to you. If reviewers asked for the changes, answer them point by point in a response to reviewers letter.