A source can pass every citation check and still be the weak point of a paper: withdrawn by its journal, a preprint cited as settled, a paper in a journal that reviews nothing, an outdated edition, or a review cited for a result it only reported.
Replacing a weak or retracted source is not filling a slot. The sentence was built on the old source, so the new one has to be read, and the sentence may change. This guide from EdCitation, which looks every reference up in the publisher's record and never writes one, gives the method and a worked example with a retracted COVID-19 study.
EdCitation's free Verify references flags retracted papers, and References from a file, part of Pro, reads your paper's whole list and recommends published sources for any reference that cannot stand. Reading the replacement stays with you.
What makes a source weak, and how is that different from a retracted one?
A weak source still stands but cannot carry the weight your sentence puts on it; a retracted source has been withdrawn by its journal and supports nothing. A weak source can often stay, with a sentence that claims less. A retracted one goes, unless the retraction is your subject.
The Committee on Publication Ethics describes retraction as correcting the literature when findings cannot be relied on, through honest error or misconduct, not as a punishment (COPE Council, 2025). Its guidelines ask for a prompt notice, free to read, that names the article, so open the notice first. The kinds of notice are explained in what a retraction is and how to check your reference list for one.
| The source | Why it is weak | Look for instead | The EdCitation step |
|---|---|---|---|
| Retracted paper | The journal no longer stands behind it | Other studies, or a review of them | Verify references flags it |
| Predatory journal | Peer review may not have happened | The finding in a reviewed journal | Find sources by the claim, then the journal's policy |
| Preprint cited as settled | Not yet peer reviewed | The published version | Find sources labels it Preprint |
| Blog or news story | A report of research | The study it describes | Find sources, by the claim |
| Outdated edition or study | Newer work may revise it | The current edition or a newer review | Find sources' Year published filter |
| Secondary source | Someone else's reading | The original, read directly | Cite a source, from its DOI |
Predatory journals
The University of Wisconsin–Madison Libraries (2026) describes predatory journals as outlets that charge authors under an open access model without rigorous peer review, and notes there is no single agreed definition. Its warning signs include vague policies on peer review and retraction, names imitating established journals, false claims of indexing or impact factors, and promises of very quick publication. If a journal does not say how it reviews papers, do not let one of them carry a claim alone.
Preprints and secondary sources
A preprint is shared before peer review, and many hold up. Brierley et al. (2022) compared early-pandemic bioRxiv and medRxiv preprints with their published versions: the conclusions of 7.2% of abstracts not about COVID-19, and 17.2% of those about it, changed in a discrete way, mostly not qualitatively. They studied only preprints that were later published. Find the published version, read it, and cite what you read.
A secondary source is a study's result as someone else reported it. APA asks you to find, read and cite the original, and to cite the secondary source only when the original is out of print, unavailable or in a language you cannot read (American Psychological Association, 2022).
How do I find out why a source is weak?
Read the record first, since it settles the retraction question, then run a library evaluation checklist.
The fullest open retraction data is the Retraction Watch Database, now distributed by Crossref, updated once per working day through its REST API and a CSV file; Crossref warns that corrections and expressions of concern are covered less completely than retractions (Crossref, 2025). It can be searched by the DOI or PubMed ID of the paper or the notice (Retraction Watch, 2024). EdCitation's Verify references runs that check on a whole list, free, and never shows a reference it could not check as not found: "not found" means the indexes answered and held no record.
Otherwise, the CRAAP test from Meriam Library (2010) at California State University, Chico, asks about currency, relevance, authority, accuracy and purpose. Three questions do most of the work: whether the information has been revised or updated, whether it was refereed, and whether another source confirms it. A claim only one weak source supports needs better support, or needs to go.
Why do retracted papers keep being cited?
Retraction does not remove a paper: it stays online, marked, and goes on being cited. Lee et al. (2021) followed a COVID-19 study that the New England Journal of Medicine retracted on 4 June 2020, a month after publishing it, amid wide publicity. Of 652 verified works citing it, 355 (54.4%) appeared three months or more after the retraction, and only 115 (17.6%) mentioned it. Seventeen fed its data into new analyses; two of those noted the retraction.
Of the 652, 36 (5.5%) were systematic reviews or meta-analyses: a review is no proof that every study inside it stands. Check the source you bring in as carefully as the one you take out.
How do I replace a weak or retracted source, step by step?
Work from the claim, not from the reference.
- Find out why the source is weak: the retraction notice, the journal's review policy, the preprint's status. Verify references flags a retracted paper; References from a file flags one inside your uploaded paper and links published corrections.
- Write down the claim: exactly what the sentence asserts, its direction, size and population.
- Decide whether the claim still stands. If the weak source was its only support, it may be wrong.
- Search for stronger support: a systematic review or meta-analysis, a newer or larger study, or the primary source. EdCitation's Find sources searches about 300 million published works by topic or by the claim a sentence needs, filters by kind of study, year published and times cited, and leaves retracted papers out unless you ask for them.
- Check the replacement's record: the published version rather than a preprint, and any corrections.
- Read it: at least the abstract, the results for your outcome and the limitations.
- Rewrite the sentence yourself to say what the new source says, with its caveats. Paraphrase, quote or summarise covers the wording.
- Cite it and check again. Cite a source builds the entry from the DOI in six styles and screens it for retraction; References from a file keeps the paper in My papers, so the next run shows what changed.
Worked example: replacing a retracted COVID-19 study
Take this sentence from a first-year essay: One large registry study found that hydroxychloroquine raised the risk of dying in hospital by about a third (Mehra et al., 2020). The paper is real: The Lancet published it online on 22 May 2020, reporting a hazard ratio of 1.335 for in-hospital death with hydroxychloroquine, from data said to cover 671 hospitals.
What the retraction notice says
Within a fortnight, Crossref's record for the paper linked a correction, an editors' expression of concern, and a retraction notice, posted on 5 June. Three of the four authors signed the notice: Surgisphere Corporation, founded by the other author, Sapan Desai, would not transfer the full dataset to independent reviewers, so the three could "no longer vouch for the veracity of the primary data sources" (Mehra et al., 2020). The paper is still online, its title now beginning RETRACTED.
What Verify references returned
We pasted the entry, in APA 7 with its DOI, into EdCitation's Verify references on 23 September 2026. It was counted as retracted, not verified, with this reason: "The DOI resolves to this record and the title matches. This paper has since been retracted." The reference was real, correctly written and still unusable.
Searching for a stronger source
For deaths in hospital, the stronger evidence is randomised trials and a pooled analysis of them. In Find sources, hydroxychloroquine mortality hospitalized COVID-19, with Systematic review and Meta-analysis ticked under Kind of study, 2021 as the first year and 50+ under Times cited, returned 130 results; the first was a meta-analysis of randomised trials by Axfors et al. (2021), cited by 305.
Ticking Randomized controlled trial instead, with 100+ citations and no year limit, put the RECOVERY trial first, but as a medRxiv preprint; searching its title found the peer-reviewed version in the New England Journal of Medicine, the one to cite. Find sources says its kind-of-study filter goes by how a paper describes itself in the title or abstract, so treat it as a first sort, not a verdict. And times cited measures attention: with retracted papers included, the retracted Lancet paper showed 1,252 citations.
Reading them and rewriting the sentence
Neither says what the old source said. The RECOVERY Collaborative Group (2020) randomised 4,716 patients in hospital: 27.0% given hydroxychloroquine died within 28 days, against 25.0% given usual care (rate ratio 1.09, 95% CI 0.97 to 1.23). Axfors et al. (2021) pooled 26 trials with 10,012 patients and found an odds ratio for death of 1.11 (95% CI 1.02 to 1.20), which they report as increased mortality, with unclear application to outpatients, children, pregnant women and people with other conditions.
Reading turned up two more points. RECOVERY and the WHO SOLIDARITY trial supply 67% of the meta-analysis's patients, so the sources are not independent confirmations. And one included trial has since been retracted; a 2024 correction reports that leaving it out keeps the odds ratio at 1.11 across 25 trials (Axfors et al., 2024). References from a file links published corrections like this; reading it tells you the claim survives.
| The sentence | |
|---|---|
| Before | One large registry study found that hydroxychloroquine raised the risk of dying in hospital by about a third (Mehra et al., 2020). |
| After | In a randomised trial of 4,716 patients admitted to hospital with COVID-19, hydroxychloroquine did not reduce deaths at 28 days (RECOVERY Collaborative Group, 2020), and a meta-analysis of 26 randomised trials found slightly higher mortality among patients who received it (Axfors et al., 2021). |
"About a third" has gone: neither replacement supports it.
What does References from a file do with a source that cannot stand?
References from a file does for every entry at once what this guide did by hand for one: reads the list out of your file, looks each reference up in the publisher's record, pairs the citations in your text with the list, and, where a reference cannot stand, recommends published sources to take its place. It belongs to EdCitation Pro and runs in your dashboard.
It is the best tool for replacing sources across a whole paper because it works from the live record, which is what carries a retraction: Lee et al. (2021) found only 17.6% of works citing one retracted study mentioned it. A chatbot asked for a replacement may write one that does not exist, as why AI tools invent references shows; References from a file recommends only real published records. It will not read them or rewrite your sentence: EdCitation never writes any part of a paper.
What you upload, what comes back, and the cost
You give it the paper, a Word document or a PDF, and it stays in My papers for the next run. Every reference gets a verdict: verified, "check this", or not found, when no publisher, registry or library holds a record of it. A withdrawn paper is flagged and a published correction linked. Citations and entries are paired both ways, so a citation left behind after a swap is caught. Where a reference cannot stand, published sources that fit are recommended, each a real record you can cite in a click. We did not run the recommendations for this guide, so they are described from the tool's own page; a whole-paper run is in upload your paper and check every reference.
Pro costs $8 a month, or $72 if you pay for the year. A free account runs References from a file three times a month; Pro has no limit. For one list before a deadline, the free Verify references does the lookups. Plans are on the pricing page, and the complete pre-submission checklist puts this check in order with the rest, Mechanics QA among them.
When should I keep a weak or retracted source?
Keep it when its weakness is your subject, or when nothing better exists and your sentence says so.
When the retraction is your subject
An essay on research integrity has to cite the retracted paper. APA 7 gives the original reference followed, in parentheses, by the year, journal, volume, issue and pages of the retraction, and a separate form for the notice, credited to whoever wrote it (American Psychological Association, 2025). Here that is three of the paper's authors:
Mehra, M. R., Ruschitzka, F., & Patel, A. N. (2020). Retraction: Hydroxychloroquine or chloroquine with or without a macrolide for treatment of COVID-19: A multinational registry analysis. The Lancet, 395(10240), 1820. https://doi.org/10.1016/S0140-6736(20)31324-6
The Lancet prints a long dash after "Retraction"; keep the title as printed.
When nothing better exists, or age is the point
A preprint on a new question may be all there is: cite it, and say it is not yet peer reviewed. A paper under an expression of concern, which COPE expects while an editor's doubts are unsettled, should not carry a claim alone. If the original is truly unavailable, name it, add "as cited in" and the work you read, and list only that work. An old edition is right when its age is the point.
Quick questions
Can I just delete a retracted reference from my paper?
Only if you rewrite or remove the sentence it supported too. Otherwise the claim is unsupported, and EdCitation's References from a file points out a citation left without an entry.
Is a preprint a weak source?
A preprint is not peer reviewed, so it is weak when cited as settled. Cite the published version when there is one, and read it first, since conclusions can change on the way.
How can I tell whether a journal is predatory?
Check its website for a stated peer-review process, a retraction policy, clear fees and real editors. Names that imitate established journals and false claims of indexing are warning signs.
Does EdCitation suggest a replacement for a retracted source?
Yes. References from a file, part of Pro, recommends published sources for a reference that cannot stand, each a real record. Free, Find sources searches by the claim. Read any replacement before you cite it.
Is a highly cited paper a strong source?
Not necessarily: a citation count measures attention. With retracted papers included, the retracted 2020 Lancet hydroxychloroquine paper showed 1,252 citations in Find sources.
References
- American Psychological Association. (2022). Secondary sources. APA Style. https://apastyle.apa.org/style-grammar-guidelines/citations/secondary-sources
- American Psychological Association. (2025). Journal article references. APA Style. https://apastyle.apa.org/style-grammar-guidelines/references/examples/journal-article-references
- Axfors, C., Schmitt, A. M., Janiaud, P., van't Hooft, J., Abd-Elsalam, S., Abdo, E. F., Abella, B. S., Akram, J., Amaravadi, R. K., Angus, D. C., Arabi, Y. M., Azhar, S., Baden, L. R., Baker, A. W., Belkhir, L., Benfield, T., Berrevoets, M. A. H., Chen, C.-P., Chen, T.-C., . . . Hemkens, L. G. (2021). Mortality outcomes with hydroxychloroquine and chloroquine in COVID-19 from an international collaborative meta-analysis of randomized trials. Nature Communications, 12, Article 2349. https://doi.org/10.1038/s41467-021-22446-z
- Axfors, C., Schmitt, A. M., Janiaud, P., van't Hooft, J., Abd-Elsalam, S., Abdo, E. F., Abella, B. S., Akram, J., Amaravadi, R. K., Angus, D. C., Arabi, Y. M., Azhar, S., Baden, L. R., Baker, A. W., Belkhir, L., Benfield, T., Berrevoets, M. A. H., Chen, C.-P., Chen, T.-C., . . . Hemkens, L. G. (2024). Author correction: Mortality outcomes with hydroxychloroquine and chloroquine in COVID-19 from an international collaborative meta-analysis of randomized trials. Nature Communications, 15, Article 1075. https://doi.org/10.1038/s41467-024-45360-6
- Brierley, L., Nanni, F., Polka, J. K., Dey, G., Pálfy, M., Fraser, N., & Coates, J. A. (2022). Tracking changes between preprint posting and journal publication during a pandemic. PLOS Biology, 20(2), Article e3001285. https://doi.org/10.1371/journal.pbio.3001285
- COPE Council. (2025). COPE guidelines: Retraction guidelines. Committee on Publication Ethics. https://doi.org/10.24318/cope.2019.1.4
- Crossref. (2025, January 19). Retraction Watch. Crossref documentation. https://www.crossref.org/documentation/retrieve-metadata/retraction-watch/
- Lee, T. C., Senecal, J., Hsu, J. M., & McDonald, E. G. (2021). Ongoing citations of a retracted study involving cardiovascular disease, drug therapy, and mortality in COVID-19. JAMA Internal Medicine, 181(11), 1535-1537. https://doi.org/10.1001/jamainternmed.2021.4112
- Mehra, M. R., Ruschitzka, F., & Patel, A. N. (2020). Retraction: Hydroxychloroquine or chloroquine with or without a macrolide for treatment of COVID-19: A multinational registry analysis. The Lancet, 395(10240), 1820. https://doi.org/10.1016/S0140-6736(20)31324-6
- Meriam Library, California State University, Chico. (2010). Evaluating information: Applying the CRAAP test [PDF]. https://library.csuchico.edu/sites/default/files/craap-test.pdf
- RECOVERY Collaborative Group. (2020). Effect of hydroxychloroquine in hospitalized patients with Covid-19. New England Journal of Medicine, 383(21), 2030-2040. https://doi.org/10.1056/NEJMoa2022926
- Retraction Watch. (2024, October 23). Retraction Watch Database user guide. The Center for Scientific Integrity. https://retractionwatch.com/retraction-watch-database-user-guide/
- University of Wisconsin–Madison Libraries. (2026, April 24). Identifying predatory journals. Where should I publish? https://researchguides.library.wisc.edu/c.php?g=1154500&p=8431826