How to Check Citations as a Peer Reviewer (Before You Sign Off)

TL;DR: A typical manuscript review has you evaluating methodology, results, and framing in a few hours — not individually looking up 60 references to confirm each one exists and actually says what the authors claim it does. That gap is exactly where citation problems survive review: reference lists padded to inflate a specific author's or journal's citation count, citations attached to the wrong year or venue, and increasingly, LLM-assisted drafting that produces plausible-looking references nobody on the author side ever double-checked. None of that shows up as a methodology problem to a reviewer working in good faith — guidelines ask you to judge whether the literature supports the argument, on the assumption that the citations themselves are real, and that assumption is exactly what breaks. Here's how to bulk-check a manuscript's full reference list against real databases in one pass, instead of spot-checking a few citations that look important and hoping the rest hold up.

The part of review that usually gets skipped

Reviewer guidelines from most journals ask you to assess whether a manuscript's claims are adequately supported by the literature it cites. In practice, that almost never means checking whether each citation is real — it means judging whether the argument built around the citations makes sense, on the assumption that the references themselves are trustworthy. That assumption is doing more work than it used to.

Citation fabrication in submitted manuscripts isn't hypothetical: padded reference lists (sometimes to inflate a specific author's or journal's citation count), citations attached to the wrong year or venue, and — increasingly — LLM-assisted drafting that produces plausible-looking references nobody on the author side double-checked. None of that shows up as a methodology problem. It shows up as a citation that doesn't hold up when you actually look for it.

Why this is worth doing systematically, not spot-checking

Spot-checking a few citations that "look important" catches the obvious cases and misses the rest — and the rest is where problems tend to hide, precisely because they weren't flagged as important enough to draw scrutiny. A padded or fabricated citation buried in a background paragraph, not the core claim, is the one least likely to get manually checked and most likely to slip through as a result. Checking the full list costs the same amount of your attention either way if it's automated; it doesn't if you're doing it by hand.

Step-by-step: verifying a manuscript's reference list

Get the manuscript as a PDF. If you're reviewing more than one submission in a cycle, or a manuscript with supplementary materials as separate files, AccuraCite's Bulk Verification plan handles up to 30 PDFs in a single scan.
Run a bulk scan from your workspace. The parser extracts the full reference list automatically — you don't need to reformat or manually re-key anything, and it handles APA, MLA, IEEE, Vancouver, and other common styles.
Every citation is cross-checked against seven databases at once — OpenAlex, Crossref, Semantic Scholar, PubMed, DBLP, arXiv, and CORE — with a consensus vote across sources rather than a single-source lookup, so an isolated bad database entry doesn't produce a false pass. Details on the How It Works page.
Scan the per-file result, not the whole PDF again. You get a verified/total ratio for the manuscript and can expand straight to the specific citations flagged as hallucinated or mismatched — the ones actually worth spending your limited review time on.
Decide what a flag means in context. A hallucinated citation in a manuscript is a different severity than one in a student essay — it can be grounds to return the manuscript to the authors for correction before review continues, particularly if the fabricated source is load-bearing for a key claim rather than incidental.

What counts as worth flagging in your review

A citation that doesn't exist at all in any indexed database — the clearest case, and one worth raising directly with the editor, not just noting privately.
A citation with the wrong year, authors, or venue — sometimes a manuscript-prep error, sometimes evidence the source was never actually consulted (a paper "cited" with the wrong claim attached to it is a common tell).
A real citation used to support a claim it doesn't actually make — verification confirms the source exists; it doesn't replace your judgment on whether it says what the manuscript claims it says. That's still on you to read.

If you're the editor handling this manuscript rather than the reviewer, checking citations belongs earlier in the pipeline too — see how AccuraCite fits into editorial review on For Publishers. And if you're grading student work rather than reviewing a submission, see how to check your students' citations for AI hallucinations.

FAQ

Does this replace my own scientific judgment as a reviewer? No — it handles the mechanical question ("does this reference exist and match its metadata?") so you can spend your review time on the part that actually requires expertise: whether the science holds up.

What if I'm reviewing under a double-blind process and don't want to expose the manuscript to a third-party tool? Treat this the same as any tool you'd use during review — check your venue's policy on third-party services before uploading unpublished manuscripts, the same way you would with any citation manager or AI-assistance disclosure requirement.

Can I check one manuscript instead of a batch? Yes — single-PDF scans are available on the Pro plan (the free tier works from pasted citation text rather than PDF upload). Bulk scanning (up to 30 PDFs per job) is for reviewers or editors handling multiple submissions in the same window, via the Bulk Verification plan or a 7-Day Bulk Pass for a single review cycle.

Should I mention citation issues in my review even if they don't change my accept/reject recommendation? Generally yes — a fabricated or badly mismatched citation is worth flagging to the editor regardless of your overall recommendation, since it's relevant to the integrity of the record even in a paper you'd otherwise accept.