To check that a citation supports its claim, read the claim, open the cited paper and ask whether the source says the same thing, something weaker, or the opposite. Two meta-analyses of medical journals found that between one in six and one in four quotations misstate their source. RefGuard, Proofig AI’s reference integrity product, flags unsupported citations automatically.
What is a quotation error?
A quotation error is a citation whose source does not say what the citing sentence says it does. Researchers who measure the problem keep it apart from bibliographic mistakes. A misspelled author name or a wrong page number is a citation error: untidy, but the reader can usually still find the paper. A quotation error is an error of content, where the citing authors misrepresent what the source reports (Baethge and Jergas, Research Integrity and Peer Review, 2025).
The expectation is written into the ICMJE recommendations that many medical journals follow. Authors are responsible for citing references accurately, and they should be able to confirm that each reference supports the statement it is attached to. A reference list can pass every formatting and metadata check and still fail that test.
How common are citations that do not support the claim?
Three reviews of medical journals point the same way:
- A 2015 systematic review and meta-analysis in PeerJ pooled 28 studies and estimated a total quotation error rate of 25.4%, with major errors at 11.9% (Jergas and Baethge, 2015).
- A 2017 review in PLOS ONE recalculated the rate per quotation examined at 14.5%. About two thirds of the content errors it counted were major: the cited source failed to substantiate the assertion, was unrelated to it, or contradicted it (Mogull, 2017).
- A 2025 update in Research Integrity and Peer Review pooled 46 studies covering about 32,000 quotations. It put the error rate at 16.9%, roughly half of it major, and found no significant improvement in quotation accuracy over recent years (Baethge and Jergas, 2025).
The exact figure depends on how each study counted errors. At these rates, a manuscript that cites several dozen sources can easily carry a citation or two that do not say what the text claims.
The 2025 authors also note that AI cuts both ways. Leaning on AI tools without checking their output, in literature searches or in writing, may add quotation errors, while AI may also help detect them.
What are the three ways a claim and its citation fail to match?
The studies above grade errors by severity. For an editor or reviewer, it helps more to sort them by what needs fixing:
- A claim that needs a source but has none. A statement of fact, a number, a comparison or a phrase such as “it is well established” appears with no reference. The reader cannot verify it. The fix is to cite the evidence, or to reword the sentence as the authors’ own interpretation.
- A cited source that does not fit. The reference is real and on topic, but it does not back this particular statement. It may study a different population, measure a different outcome, report an association where the text claims a cause, or mention the subject in passing. A review cited in place of the original study often lands here too: ICMJE asks authors to cite original research directly whenever possible, because reviews may not reflect the original work accurately.
- A cited source that contradicts the claim. The paper reports the opposite, a null result, or a finding the citing text reverses. This is the most damaging case: the reference hands the reader evidence against the sentence while appearing to support it.
Single errors add up. In a 2009 BMJ analysis, Steven Greenberg traced 242 papers and 675 citations around one biomedical claim. He showed how citation bias, amplification and invention, including turning a hypothesis into an accepted fact through citation alone, gave the claim an authority its data did not support.
How do reviewers and editors check citations by hand?
They sample. The 2015 meta-analysis lists spot checks by editors and reviewers among the measures against quotation errors, together with placing citations correctly in the text and asking authors to declare that they checked the cited material. A practical spot check works like this:
- Pick the claims that carry the argument. Numbers, causal statements and sentences that justify the study design are where a misquoted source does the most damage.
- Match each claim to its citation. Check that the reference sits next to the statement it supports, not at the end of a paragraph that makes three different points.
- Open the cited paper. Read the abstract, then the relevant results. Note the population, the outcome and the direction of the effect.
- Classify the match. Supports, partly supports (narrower or weaker than the text implies), does not address, or contradicts.
- Scan for uncited claims. Read the introduction and discussion again, looking for statements of fact with no reference at all.
- Follow the chain. Where a review stands in for a primary study, ask for the original source.
This works, and it is slow. A reviewer who checks five citations in a reference list of fifty has checked five, and the rest of the list is taken on trust.
How can RefGuard help you check claims against citations?
RefGuard, Proofig AI’s reference integrity product, checks claims against citations across the whole manuscript. It flags two things: statements that need a source but lack one, and citations that do not align with the claim, including cited sources that contradict it. RefGuard shows every flag with the claim in context, so a person can open the source and decide.
The same pass checks the reference list itself:
- Existence. References that cannot be matched to any real publication are flagged, and this is where hallucinated citations from AI writing tools surface. The Octym blog explains how fabricated references get into papers.
- Retractions. Citations to retracted papers are flagged with the reason, the date and the type of notice. Another Octym post covers why a reference can be retracted after you cite it.
- Incomplete or broken references. RefGuard points out entries with a wrong year, a missing author or other broken details, so they are easy to fix.
RefGuard accepts text-based PDF and Word files of up to 150 pages. It is available on its own, inside PubShield for publishers and research institutions, and as the reference checks in Octym, which reviews the entire paper with 26 checks. Editors can use it to screen submissions, and authors can use it to check their references before they submit.
Results are signals for a person to weigh, not verdicts, and authors remain responsible for their references. What changes is coverage: the reviewer no longer samples five citations at random, but starts from the ones that most need a human look. See what RefGuard checks in a manuscript.
Frequently asked questions
What is the difference between a citation error and a quotation error?
A citation error is a mistake in a reference’s bibliographic details, such as a misspelled author, a wrong year or an incorrect page range. A quotation error is a mistake in content: the cited source does not say what the citing sentence claims. Citation errors can make a paper harder to find, while quotation errors mislead the reader. RefGuard helps with both: it points out incomplete or broken references, and it flags citations that do not support the claims they are attached to.
Is citing a review article instead of the original study a mistake?
Citing a review article instead of the original study is a mistake when the claim rests on a specific finding. The ICMJE recommendations ask authors to cite original research directly whenever possible, because reviews may not reflect the original work accurately. For an effect size or a trial result, cite the study that reported it. A review is the right reference when the claim is about the overall state of the evidence, for example that results in a field are mixed. RefGuard, Proofig AI’s reference integrity product, flags unsupported citations.
Who is responsible for checking that a citation supports its claim?
Authors are responsible for checking that each citation supports its claim. The ICMJE recommendations hold them accountable for citing references accurately and expect them to be able to confirm that each reference supports its statement. Editors and reviewers add spot checks, which a 2015 meta-analysis lists among the measures against quotation errors. Automated checks such as RefGuard from Proofig AI widen that net, but they produce signals for review, and the decision to keep, replace or reword a citation stays with the authors.
