Citation Verification for Journal Editors: How to Catch Bad References Before Peer Review

As an editor, you've seen the pattern: a well-written manuscript arrives with a reference list that looks professional at first glance. Thirty or forty citations, properly formatted, covering the relevant literature. You send it to reviewers. Three weeks later, a reviewer flags that reference 17 doesn't exist and reference 23 cites a retracted paper. Now you've wasted reviewer time, damaged credibility, and created an awkward exchange with the authors.

In 2026, this scenario is more common than ever. AI writing assistants have made it trivially easy for authors — even well-intentioned ones — to include references that are fabricated, distorted, or outdated. The problem isn't just with bad actors. It's with a publishing ecosystem where generating a plausible-looking citation is easier than verifying one.

This guide is for editors who want to add a reference screening step to their desk review workflow without doubling their workload.

The Scale of the Problem

Multiple studies have documented the rise of fabricated references in submitted manuscripts:

• A 2025 analysis of 500 manuscripts submitted to a mid-tier medical journal found that 12% contained at least one reference that could not be verified against CrossRef or PubMed.
• Retraction Watch reported a 340% increase in retractions citing "unreliable references" between 2023 and 2025.
• Editors at several computer science venues reported that manuscripts generated with AI assistance had fabricated reference rates between 15% and 30%.

These numbers represent manuscripts that made it through the authors' own review. The actual generation rate of fake citations by AI tools is much higher — the ones that reach your desk are the ones the authors didn't catch.

What Editors Can Realistically Check at Desk Review

You're not expected to verify every reference in every submission. But a targeted screening catches the most damaging problems:

Tier 1: Automated batch verification (2 minutes per manuscript)

Copy the reference list into an automated checker like Citely's Citation Checker. The tool parses each reference and verifies DOIs against CrossRef. This single step catches:

• DOIs that don't resolve (fabricated or mistyped)
• DOIs that point to a different paper than described
• Missing DOIs that should exist
• Major metadata discrepancies (wrong year, wrong journal)

For a 40-reference manuscript, this takes about 60 seconds of processing time. The total workflow — copy, paste, review results — is under 2 minutes.

Tier 2: Retraction check (1 minute per manuscript)

Cross-reference the DOIs against retraction databases. CrossRef metadata includes retraction notices for many publishers. The Retraction Watch database is another source.

A cited retracted paper is one of the most serious integrity issues you can catch at desk review. It's also one of the easiest to miss, because the paper existed and was valid when originally published.

Tier 3: Spot-check high-risk references (5 minutes per manuscript)

Some references deserve manual attention:

• References to papers published in the current year — more likely to be fabricated, since databases may not have indexed them yet
• References with round publication years (2020, 2015) — AI tools show a slight bias toward round years
• Self-citations — verify they point to real publications by the authors
• References to obscure or predatory journals — check that the journal exists and is indexed

Building This Into Your Editorial Workflow

For journals with editorial assistants

Train your editorial assistant to run every submission through automated verification before you see it. The assistant flags manuscripts with issues and includes the verification report. You make the editorial decision based on the severity:

• 1-2 minor metadata errors → Note in the decision letter, ask authors to correct
• Fabricated references → Desk reject with explanation
• Retracted paper cited → Flag for integrity review

For editors handling everything themselves

Add a 3-minute step after your initial read-through and before you decide whether to send to reviewers:

1. Copy the reference list (60 seconds)
2. Run automated check (60 seconds of processing)
3. Scan the report for red flags (60 seconds)

This small time investment prevents the much larger time cost of reviewer complaints, author correspondence, and potential corrections after publication.

For editorial boards setting policy

Consider adding reference verification to your submission guidelines:

"Authors are expected to verify all references against CrossRef or equivalent databases prior to submission. Manuscripts containing references that cannot be verified may be returned without review."

This shifts some responsibility to authors while signaling that your journal takes citation integrity seriously.

Red Flags in Reference Lists

Beyond what automated tools catch, experienced editors develop an eye for these patterns:

Suspiciously uniform formatting. If every single reference is perfectly formatted in exactly the same style, with no variations or quirks, the list may have been generated rather than compiled over time. Real reference lists assembled during months of research usually show small inconsistencies.

All references are highly relevant. Real literature reviews include tangential and background references. If every citation is a perfect keyword match for the paper's topic, the references may have been generated to fit rather than discovered through research.

Author names that don't match the field. A paper on computational linguistics citing a "Smith et al." in the Journal of Computational Linguistics — but the only Smith who publishes in that journal works in a completely different subfield. This pattern is characteristic of AI-generated chimera references.

Clustering of references from 2020 or 2023. AI training data has natural cutoff points, and generated references tend to cluster around these dates.

Key Takeaways

• 12% of submitted manuscripts in a recent study contained at least one unverifiable reference — most from AI writing assistance, not intentional fabrication
• Automated batch verification takes under 2 minutes per manuscript and catches fabricated DOIs, metadata errors, and retracted papers before they waste reviewer time
• A three-tier screening system (automated check → retraction check → spot-check high-risk references) fits within a desk review workflow without adding significant time
• Consider adding reference verification requirements to submission guidelines — it shifts responsibility to authors and signals editorial standards
• The highest-risk references to spot-check manually are current-year publications, round-year publications, and citations to obscure journals

Screen your submissions → citely.ai/citation-checker