To check a paper before you cite it, find the sentence in the full text that supports your claim, confirm the study design can carry that claim, and confirm the population matches your own. Then check the article page and its Crossref record for a correction or retraction, and record where the claim sits.
Citation errors spread because references are copied, not read
A citation error is a reference attached to a statement it does not support. Checking the format of a reference does not test for it. Three mechanisms produce it.
- Abstract-only reading. An abstract gives the headline result without the comparator, the denominator or the caveats that qualify it. A sentence built on it carries none of those qualifiers.
- Chained citation. You read paper B, which cites paper A, and you cite A without opening it. If B misread A, you have reproduced the error and added a second source to it.
- Design drift. The source reports an association; your sentence says the exposure reduced the outcome. The number is real. The verb is not.
How to check a paper before you cite it: five steps
Run these five checks on every reference you intend to keep.
| Step | What you are testing | Where to look | Fails when |
|---|---|---|---|
| 1. Locate the claim | A sentence in the source states your claim | Results, table or figure. Not the abstract | The support sits only in the source's introduction |
| 2. Test the design | The design can carry the verb in your sentence | Methods: design, comparator, outcome measurement | Your sentence claims cause; the design shows association |
| 3. Match the population | The people, setting and period transfer to yours | Methods: eligibility criteria, site, dates | A specialist-centre finding is applied to unselected patients |
| 4. Check the status | The article is not corrected or retracted | The article page and the Crossref record | You are working from a PDF saved before the notice appeared |
| 5. Record the location | The support can be found again | Your citation record, written as you verify | Months later, nobody recalls why the reference is there |
Step 1 is the step the other four depend on. Copy the supporting sentence out of the source and paste it beside your own. Where there is no supporting sentence, the reference is decoration.
Match the verb in your sentence to the design of the study you cite
Your verb sets the claim, not the standing of the source. A large study in a prominent journal cannot support a causal verb if its design was observational. Judging a source by the journal it appeared in, or by the metrics attached to it, is a separate error; it helps to know what those metrics actually measure.
| If your sentence says | The source has to be | Frequent mismatch |
|---|---|---|
| X reduces or causes Y | A randomised trial, or a stated causal design | A cross-sectional study reporting a correlation |
| X is associated with Y | An analytic study with a comparison group | A case series with no comparator |
| X occurs in a given proportion | A defined population with a real denominator | A referral-centre series, where the denominator is those referred |
| X is safe | Powered and long enough to see the harm you exclude | A short trial reporting no serious events in a small sample |
Change the verb rather than hunting for a more convenient source.
Check the population and setting before you generalise the finding
A finding transfers only as far as the population it was measured in. Before citing a result as though it applies to your patients, read the eligibility criteria and record four things.
- Who was studied. Condition, severity, and how participants entered the study.
- Where. Primary care, a district hospital or a specialist referral centre, and in which country.
- When. The recruitment dates, and whether practice has changed since.
- Who was left out. The exclusions are the informative part: a trial that excluded people with renal impairment cannot tell you what happens in renal impairment, however large it was.
Check correction and retraction status while you are still building the list
A retraction is a formal declaration by the journal or the authors that a published article should no longer be relied on, whether because the findings are unreliable or because the work breached research or publication ethics. The article normally stays online, marked with a retraction notice, rather than disappearing from the record.
- The publisher's article page. Open the version of record online, where a correction, an expression of concern or a retraction notice should be linked from the article.
- The Crossref record. Update relationships travel in the DOI metadata when the publisher deposits them. Crossref exposes them at
https://api.crossref.org/works/{DOI}, where a deposited correction or retraction appears as a linked update. A PDF saved on your own machine carries no such link. A missing link is not proof that no notice exists. - Read the notice itself. A partial retraction, a correction and an expression of concern mean three different things. The NISO CREC recommended practice, which covers retractions, removals and expressions of concern, exists because those notices have been inconsistently labelled and slow to travel.
- Preprints. Check whether the preprint you cite has since been published, revised or withdrawn, and cite the version of record.
If a paper you relied on is retracted, the argument that rested on it has to be rebuilt, not re-referenced.
Record where the claim sits so it can be rechecked later
Verification you did not write down has to be repeated. Keep one row per reference, filled in as you check it, so a co-author or reviewer can follow your reasoning.
| Field | What you record | Worked example |
|---|---|---|
| Your sentence | The claim as it stands | Delayed diagnosis was associated with longer inpatient stay. |
| Source | Author, year and DOI | DOI copied from the publisher page |
| Where the support sits | Section, paragraph, table or figure number | Results, third paragraph; Table 2, adjusted model |
| Design and size | Design, sample size, centres | Retrospective cohort, 412 patients, one tertiary centre |
| Population | Who, where, when, and key exclusions | Adults admitted 2016 to 2019; children excluded |
| Status checked | What you checked, and when | Article page and Crossref record, no notices, 12 March |
| Verdict | Supports as written, supports less, or not at all | Association only; verb changed from reduced to was associated with |
A reference manager records how a citation is formatted, not whether the source supports your sentence; the format is a house rule, set out in our comparison of APA, Vancouver and MLA styles.
What to do when the source does not support your sentence
Finding a mismatch is a success. There are four honest responses, in order of preference.
- Weaken the claim to what the source shows. Change caused to was associated with, or all patients to the patients studied. It costs you one sentence.
- Cite the primary source instead. If you met the claim in a review, trace it to the study that produced it and read that study.
- Attribute the claim to its origin. Where a statement is a proposal rather than a finding, say so: one group has proposed X, rather than X is the case.
- Delete the reference. A sentence propped up by an irrelevant citation is worse than an unsupported one: it looks settled.
For reviewers: how to spot-check a reference list
A reviewer cannot verify every reference, but spot-checking the load-bearing ones is realistic.
- Identify the sentences the conclusion depends on: the end of the introduction, the rationale for the method, the opening of the discussion.
- Open the references attached to those sentences and find the supporting statement in the full text.
- Compare the verb to the design: a causal verb needs a randomised or otherwise causal design, and an association verb needs a comparison group.
- Check the status of any reference that is central to the argument.
- Write specific comments. Reference 14 is a case series and cannot support a causal claim is actionable; the referencing is poor is not.
The ICMJE Recommendations place the accuracy of references on the authors, who are asked to verify them against the original documents and not to cite retracted articles except as retractions. COPE guidance approaches citation problems from the editorial side, and asks reviewers to raise concerns they notice with the editor.
What we check in your reference list, and what we do not
Directive Publications does not publish a reference-accuracy policy, and we do not verify every citation in every manuscript. Similarity screening runs on submissions and compares text, so it finds a copied passage but never a misattributed claim. Peer review here is double-blind, and reviewers may raise citation problems they notice, but the reference list is not re-read for you here. Verification is the authors' work.
The standards this sits under are set out in our research integrity policy and our publication ethics policy, both aligned with the COPE Core Practices. Read next our guides to what a retraction means for the citing author, types of research article and citing data, software and code.