A reviewer judges the methods and the argument, and the reference list often gets a glance at the end. Yet it holds some of the easiest problems to prove: a withdrawn paper cited as evidence, an entry no database knows, a run of citations to one journal that the text never discusses. This guide covers how to peer review a paper's references and how to write it up for the editor.
EdCitation publishes this guide. Every rule below was read at its source on 27 September 2026: the Committee on Publication Ethics (COPE) and the reviewer guidance of Elsevier, Springer Nature and Wiley. EdCitation's free Verify references looks each entry up in the publisher's record and flags retractions; we ran it on six test references and quote what came back. It looks things up and never writes a word of anyone's work; the judgement is yours.
How refereeing became standard is in the history of peer review; the author's side is the journal manuscript references checklist.
What does a peer reviewer have to check in the references?
Two things: whether the list serves the paper honestly, and whether each entry is what it claims to be.
What COPE asks of reviewers
COPE's Ethical Guidelines for Peer Reviewers (version 2, September 2017) set four duties that bear directly on references (COPE, 2017):
- Back your criticism. Be specific, and support general statements with evidence and "appropriate references".
- Do not farm citations. Do not suggest that authors cite your own or an associate's work merely to raise its citation counts or visibility; any suggestion needs a valid academic or technological reason.
- Report, do not investigate. If you suspect misconduct in the research or its writing, tell the editor and "do not attempt to investigate on your own".
- No accusations in the confidential box. Comments meant only for the editor are not a place for denigration or unfounded accusation.
What the publishers add
Elsevier's reviewer page asks you to scrutinise whether the references are "the most up-to-date and relevant" for the topic, and to raise suspected plagiarism, falsification or other ethical concerns with the editor, "providing as much detail as possible" (Elsevier, n.d.). For review articles, recommend the foundational studies or recent publications the list should include. The same page says to keep comments strictly factual and not to speculate on the authors' motives.
Springer Nature's guidance says to watch for missing references, to check that those cited are still current, and prefers original studies to review articles as sources (Freedman, 2016).
Peer review checklist for references, in one table
This peer review checklist puts each check next to how to do it, what to write, and the rule behind it. Where the journal has its own form, use it, as COPE (2017) asks.
| What to check | How | What to write to the editor | Source |
|---|---|---|---|
| Key work missing | Read the introduction and discussion against what you know; confirm a paper you would add exists in Find sources | To the authors: name the paper and the claim it bears on | Elsevier (n.d.); Freedman (2016) |
| Excessive self-citation or citation stacking | Count citations to the authors' own work and to one journal; ask whether each is discussed | To the editor: the count and the reference numbers, without a motive | COPE (2019); Wiley (n.d.) |
| Retracted work cited as sound | Look for a retraction notice at the publisher; run the list through Verify references | To the authors: the reference, the notice and its year | Elsevier (n.d.) |
| Reference that cannot be found | Search by DOI, title and author; a failed lookup is not proof | To the editor: which entries, where you searched, a request for DOIs | COPE (2017) |
| Source that does not support the claim | Open the cited work and read the passage beside the sentence | To the authors: line number, what the source says, what the text says | COPE (2017) |
| Wrong year, author or volume | Compare with the publisher's record; Cite a source shows the record's details | To the authors, as a minor point | Elsevier (n.d.) |
| Your own work, if you are tempted to add it | Would you suggest it if someone else had written it? | Only with the scholarly reason stated | COPE (2017); Elsevier (n.d.) |
How do I check the references in a manuscript, step by step?
Work cheapest first, so that the slow reading goes where the fast checks point.
- Read the journal's reviewer instructions and its policy on outside tools first.
- Skim the list for shape. How many entries, how recent, how many by the authors themselves, how many from one journal.
- Look every entry up. With the editor's agreement, paste the list into EdCitation's Verify references: each entry comes back verified, doubtful ("check this") or not found, and retracted papers are flagged. Without that agreement, look up by hand the entries the argument leans on.
- Open the five or six references that carry the main claims and read the cited passage beside the sentence that cites it.
- Note what is missing, with the full reference of each paper you would add.
- Sort what you found: corrections for the authors, concerns for the editor.
- Write it up factually, with reference numbers and line numbers.
Step 4 is where reviewing time is best spent: a source that does not say what the text claims misleads a reader, and no database lookup can find it. How often that happens is in citation errors in published research.
How do I spot citation manipulation and coercive citation?
Citation manipulation is citing work to raise its citation counts rather than because the paper needs it. COPE's 2019 discussion document names three patterns: excessive self-citation by an author, excessive citation of the journal the paper is submitted to, and excessive "honorary" citation or "citation stacking" to benefit another author or journal (COPE, 2019). The test is whether the references "contribute to the scholarly content" of the article.
Self-citation and citation stacking in a manuscript
Self-citation is not wrong in itself. COPE's 2024 position says it can be legitimate, for example in a small field, and leaves thresholds to each journal's policy (COPE, 2024). What a reviewer can usefully report is the pattern: a block of self-citations on one sentence, references to one journal the text never discusses, several citations piled onto one point, which Wiley's page for editors lists among its signs (Wiley, n.d.). Report the count and the reference numbers; the editor, who knows the journal's policy, decides what it means.
Coercive citation: when the pressure comes from reviewers
Coercive citation is citation made a condition of acceptance. COPE (2019) says it can come from reviewers who ask for their own articles to be cited when the editor turns the request into a requirement, and from editors who require citations to their own work or journal. Elsevier and Wiley carry the same prohibition for reviewers. Elsevier (n.d.) lists suggesting your own or your associates' work, except for genuine scientific reasons, among the things not to do. Wiley (n.d.) says reviewers should not ask for citations without a strong scholarly rationale, and tells editors that a report made up mostly of self-citation requests may be set aside.
If your own paper really is the one the authors need, say why in the confidential comments and let the editor decide.
What if a reference is fake, retracted or does not support the claim?
Each of the three calls for a different response, and none of them calls for an accusation.
A reference that cannot be found
A reference nobody has a record of may be invented, the kind of fake reference AI chatbots produce, or it may be a mistyped title, a report in another language or a thesis outside the big indexes. The evidence on invented references reaching accepted papers is in what studies say about AI-fabricated references. Your task is narrower: say which entries could not be found and where you searched, and ask for a DOI or a link. COPE (2017) says not to investigate further unless the journal asks.
A lookup that failed because a service did not answer is not evidence that a work does not exist. EdCitation's Verify references never shows "could not check" as "not found", and a report should draw the same line.
A retracted paper cited as if it were sound
Cited as evidence, a retracted paper leaves the claim without support. How to check one, and how a retracted entry should be marked, is in retracted papers in your reference list. Name the reference, the notice and its year, and ask the authors to replace the source or explain why it stays.
A source that does not support the sentence
Only a reader can make this check. Open the cited paper, find the passage, and write what the source reports beside what the manuscript says it reports, with the line number: the specific, evidenced criticism COPE (2017) asks for.
How do I report a reference problem to the editor without accusing anyone?
Describe what you found, where, and how you checked, and leave out any guess about why. Elsevier (n.d.) says not to speculate on motives; COPE (2017) keeps accusations out of the confidential comments too. A slip and misconduct can produce the same reference list, and telling them apart is the editor's job.
Example wording, to adapt to your journal:
- For the authors: "Reference 23 was retracted in 2010 (the notice is linked from the publisher's page). Please replace it, or cite it as retracted if the retraction is relevant to the argument."
- For the editor: "I could not find records for references 7 and 31 in Crossref, PubMed or Google Scholar, searching by title and author. It may be a citation error; it would help if the authors could supply a DOI or link for each."
- For the editor: "Eleven of the 48 references are to the authors' earlier work, and references 12 to 19 are cited together in one sentence without discussion. I raise it only so that you can judge it against the journal's policy."
Each states a fact, a place and a request. None says "fabricated", "fraud" or "manipulation": those are conclusions for the editor, if anyone reaches them.
Can I put a manuscript's references into a checking tool?
Only as the journal allows. A manuscript under review is confidential: COPE (2017) says not to involve anyone else without the journal's permission, and Elsevier (n.d.) says the materials cannot be shared without the editor's prior authorisation.
The publishers' rules on software are explicit. Elsevier says reviewers should not upload a manuscript "or any part of it" into an AI tool, and allows AI only in a supporting role, in private tools, with disclosure (Elsevier, 2026). Springer Nature says reviewers must not upload manuscript content to unsecured or public AI tools, must disclose AI use, and must not delegate judgement to AI (Springer Nature, n.d.).
A reference list is part of the manuscript. So ask the handling editor before pasting the list, or uploading the file, into any outside service, EdCitation included, and follow the answer.
What Verify references said about six test references
We pasted six references on peer review and citation practice into Verify references on 27 September 2026: one retracted paper, one wrong year, one wrong author, one invented entry, and two correct ones (the second, Smith's 2006 essay without a DOI, was also verified). The verdicts:
- Schroter and colleagues' 2008 study of error detection by BMJ reviewers, with its DOI: verified, "The DOI resolves to this record and the title matches." Its abstract reports that 607 reviewers, sent test papers with nine major errors each, found 2.58 of them on average at baseline (Schroter et al., 2008): a reason to work from a checklist.
- Wakefield and colleagues' 1998 Lancet paper, without a DOI: verified and flagged, "Title, year and first author match a published record. This paper has since been retracted." The record returned lists a 2004 correction and a 2010 retraction notice.
- Wilhite and Fong's coercive citation paper in Science, which we dated 2014: doubtful, "The DOI and title are right, but the work was published in 2012, not 2014."
- Fong and Wilhite's 2017 PLOS ONE paper, with the second author changed to an invented "Whitehead": doubtful, "The publisher's record also names Wilhite, who is not in your entry. Check the authors."
- "Harrow and Delaney (2021)", a paper we invented for the test: doubtful, "The closest published record differs in its title. Check it is the one meant, and copy the title from it." The closest record it offered was a 1990 JAMA paper by a different author, in a different journal and year.
Reading those verdicts as a reviewer
The retraction and the two detail errors came back as a reviewer would want: one line each, with the correction to ask for. The invented reference was not verified, but came back "check this" rather than "not found", because a paper on a similar subject exists; the offered match differs in author, journal and year, which takes seconds to see. The tool narrowed six entries to four that need a human look, and the reviewer decides what each means.
Which EdCitation tool fits a reviewer's reference check?
For confirming that each entry is a real, published work and has not been withdrawn, EdCitation's Verify references is the best tool we know of: it looks every reference up in publishers' and registries' records and never writes one, where a chatbot can invent the answer. In the run above, a retraction, a wrong year and a missing co-author each came back with the correction attached.
- Verify references, free with no account: paste a list or upload a paper; each entry comes back verified, doubtful ("check this") or not found; retracted papers are flagged; "could not check" is never shown as "not found". With a whole paper uploaded, every in-text citation is matched to the list and every entry back to the text.
- References from a file, part of Pro at $8 a month: reads a whole manuscript file, looks every reference up, matches every in-text citation to the list, and links published corrections. Three checks a month come with a free account (pricing).
- Cite a source, free: builds an entry from a DOI, PubMed ID, title, ISBN or web address in APA 7, MLA 9, Chicago 18 author-date, Harvard, IEEE or Vancouver, screened for retraction, for handing the authors a paper you recommend.
EdCitation writes no part of anyone's work, a review included: every judgement stays the reviewer's own, as COPE and the publishers require. Upload a manuscript only where the journal allows it.
Quick questions
Should a peer reviewer check every reference?
No guidance we read requires every entry to be checked by hand. Elsevier asks reviewers to review the references carefully; a lookup, where the journal allows it, covers every entry, and the reading goes to the references carrying the main claims.
Can I ask the authors to cite my own paper?
Only for a genuine scholarly reason, stated in the report. COPE's guidelines and Elsevier's reviewer page both rule out suggesting your own work to raise its citations, and a request made a condition of acceptance is coercive citation.
What should I do if a reference in a manuscript looks fake?
Tell the editor which entries you could not find and where you searched, and ask for a DOI or link. Do not call it fabricated; COPE tells reviewers to report concerns and not to investigate on their own. EdCitation's Verify references shows which entries have no record.
Can I upload a manuscript under review to an AI tool?
Not without the journal's permission. Elsevier says reviewers should not upload a manuscript or any part of it into an AI tool, and Springer Nature rules out unsecured or public AI tools. Ask the editor before using any outside service.
References
- Committee on Publication Ethics. (2017). COPE ethical guidelines for peer reviewers (Version 2). https://doi.org/10.24318/cope.2019.1.9
- Committee on Publication Ethics. (2019). COPE discussion document: Citation manipulation. https://doi.org/10.24318/cope.2019.3.1
- Committee on Publication Ethics. (2024). Handling citation manipulation [COPE position]. https://doi.org/10.24318/raEKBDOK
- Elsevier. (n.d.). How to conduct a review. https://www.elsevier.com/reviewer/how-to-review
- Elsevier. (2026). Generative AI policies for journals. https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals
- Freedman, P. (2016, September 23). 10 steps to evaluating manuscripts as a peer reviewer. Springer Nature. https://www.springernature.com/la/researchers/the-source/blog/blogposts-for-peer-reviewers/10-steps-to-evaluating-manuscripts-as-a-peer-reviewer/16679144
- Schroter, S., Black, N., Evans, S., Godlee, F., Osorio, L., & Smith, R. (2008). What errors do peer reviewers detect, and does training improve their ability to detect them? Journal of the Royal Society of Medicine, 101(10), 507–514. https://doi.org/10.1258/jrsm.2008.080062
- Springer Nature. (n.d.). Artificial intelligence (AI). Nature Portfolio editorial policies. https://www.nature.com/nature-portfolio/editorial-policies/ai
- Wiley. (n.d.). Identifying citation manipulation. Editor resources. https://www.wiley.com/en-us/publish/article/editor-resources/maintaining-ethical-practices/identifying-citation-manipulation/