Pre-Submission Reference Audit Checklist for Journal Authors

A single citation error can trigger desk rejection. The difference between acceptance and immediate rejection often comes down to whether your reference list survives the editorial office’s first five-minute scan.

This isn’t paranoia. Desk rejection due to formatting or citation errors has been one of the most common reasons papers never see peer review since at least 2020, and the problem only grew worse with AI-generated hallucinated citations. The Lancet/CITADEL audit found that 1 in 277 papers published in early 2026 contained a fabricated reference — a twelve-fold increase from 2023 — and editors know it. Many now deploy automated screening that flags citation problems before any human reviewer ever opens your manuscript.

This article gives you a complete, copyable pre-submission reference audit checklist. Run it before you submit, and you’ll catch the errors that cause desk rejection, revision requests, or — worse — post-publication corrections and retractions.

What Journal Editors Actually Check (And What Gets Papers Desk-Rejected)

The editorial office doesn’t have time to read your paper. When a submission arrives, a desk editor or production staff runs through a technical screening checklist that takes five to ten minutes. Reference issues are among the most common problems they find:

• Mismatched in-text citations: References cited in the body but missing from the reference list, or vice versa
• Inconsistent formatting: Random citation styles (APA mixed with Vancouver), inconsistent capitalization, missing DOIs
• Non-existent or unverifiable references: AI-hallucinated citations, retracted papers presented as valid, duplicate citations numbered differently
• Journal abbreviation errors: Using full journal names where abbreviated names are required (or the reverse), incorrect ISO 4 abbreviations
• Wrong volume/issue/page numbers: Metadata that looks correct but doesn’t match the actual publication

ReferenceChecker.org (2026) catalogued ten distinct citation error categories that trigger desk rejection or revision requests, and the list is growing as journals adopt automated verification tools. COPE now recommends that member journals implement automated reference screening at submission because “manual checking does not scale” in high-volume environments.

The Three-Layer Audit Framework

The core insight of this article is that “citation checking” isn’t a single task. It’s three distinct jobs that most tools only partially address:

A proper reference audit addresses three distinct layers. Most tools only cover the first two — Layer 3 (external verification) is what catches hallucinated and fabricated citations. The three-layer framework is visualized below:

A proper audit addresses all three layers. If you only do style formatting (Layer 1) or only check for existence (Layer 3), you’ll miss errors that fall in between. Here’s what each layer means and how to audit it.

Layer 1 — Style Conformance

What to check: Every reference must follow the journal’s citation style exactly.

• Citation format matches the journal’s guide (author-year, numbered, Vancouver, APA, IEEE, Chicago, etc.)
• Punctuation, capitalization, and italics conform to the style guide
• Journal abbreviations follow ISO 4 standards where required
• DOI URLs are formatted correctly and lead to valid pages
• Reference list ordering matches the citation style (alphabetical, numerical, etc.)

Tools: Reference managers (Zotero, EndNote, Mendeley), style-specific plugins, or manual compliance against the journal’s author guidelines. Most reference managers handle this layer automatically once you select the correct style.

What this catches: Formatting inconsistencies that desk editors spot in seconds. These are the easiest errors to make and the most embarrassing to have caught — they signal carelessness rather than intellectual problems.

Layer 2 — Structural Consistency

What to check: The internal logic of your citation system must be coherent.

• Every reference in the reference list appears at least once in the body text
• Every citation in the body text appears in the reference list
• No duplicates: two entries for the same paper with slightly different formatting
• No ghost citations: in-text citations to papers that don’t exist in the reference list
• Consistent numbering or author-year format throughout (no mixed systems)
• All co-authors listed according to journal policy (many limit to 6-10 names then use “et al.”)

Tools: Manual cross-check, citation checker tools (Citely, Paperpal), or automated reference management exports.

What this catches: The “I cited something but forgot to add it” errors that happen when you revise text but don’t update the reference list, and vice versa.

Layer 3 — External Verification

What to check: Every reference must be real, correctly attributed, and still valid. This is the hardest layer and the one AI-hallucinated citations bypass.

• Resolve every DOI at doi.org manually — don’t trust embedded links (they can redirect to wrong papers)
• Search every title in at least one independent database (Crossref, OpenAlex, Google Scholar) to confirm the paper exists and matches the citation metadata
• Check for retraction status using Retraction Watch, Crossref, or Zotero’s retraction flags
• Verify authors actually work in the cited discipline (AI often pairs real names from adjacent fields)
• Confirm the paper’s claims support what you’re citing them for — real papers, wrong claims

Tools: DOI resolver (doi.org), Crossref Simple Text Query, OpenAlex, PubMed, Retraction Watch Database, Scholar Sidekick, answerThis.io. Zotero’s built-in retraction notifications cover approximately 75% of papers (items with DOI or PMID); the remaining 25% need manual checking.

What this catches: AI-hallucinated citations, misused sources (real paper, wrong claim), and papers that have been retracted since you last checked. The Columbia University 2.5M paper audit found that even among “real” AI-generated citations, 43% of GPT-3.5 outputs and 24% of GPT-4 outputs contained substantive errors — wrong year, wrong author, or wrong journal. A “real” DOI does not mean accurate metadata.

The Pre-Submission Checklist (Copyable)

Here’s the complete checklist. Run through every item before you submit. The process should take 30–90 minutes depending on the number of references.

Pre-Submission Reference Audit Checklist

A. Style Conformance

• [ ] Confirm the journal’s required citation style (check the author guidelines)
• [ ] Verify every citation follows the journal’s formatting rules (punctuation, capitalization, italics)
• [ ] Check journal abbreviations against ISO 4 (e.g., “Journal of Clinical Investigation” → “J. Clin. Invest.”)
• [ ] Confirm DOI format is correct and complete for every reference
• [ ] Verify reference list ordering (alphabetical, numerical, etc.) matches the citation style
• [ ] Check that author names are formatted correctly (initials vs. full names, depending on style)

B. Structural Consistency

• [ ] Cross-reference: every citation in the body appears in the reference list
• [ ] Cross-reference: every reference in the list is cited in the body text
• [ ] Check for duplicate entries (same paper listed twice with different formatting)
• [ ] Confirm co-author limits match journal policy
• [ ] Verify in-text citation format matches reference list format consistently
• [ ] Check that “et al.” usage is consistent throughout (check journal policy on first mention vs. subsequent mentions)

C. External Verification

• [ ] Resolve every DOI at doi.org manually; confirm it leads to the correct paper
• [ ] Search every title in at least one independent database (Crossref, OpenAlex, Google Scholar)
• [ ] Check each reference for retraction status (Retraction Watch, Zotero, PubMed)
• [ ] Verify authors actually work in the cited discipline (not adjacent fields)
• [ ] Read the abstract of each source to confirm the claim you’re citing is actually supported
• [ ] Confirm publication year matches the citation (AI frequently mixes up dates)
• [ ] If you used AI tools during writing, assume every citation is hallucinated until verified

D. Final Compliance

• [ ] Run your reference list through Crossref Simple Text Query (free, up to 1,000 references)
• [ ] Check for “vibe citations” — references that look correct on surface but contain errors on inspection (real authors, wrong paper)
• [ ] Verify that preprints you cited have been updated to their peer-reviewed versions if available
• [ ] Confirm the reference manager’s export matches your formatted reference list (no formatting loss during export)
• [ ] Run one final read-through: does the reference list tell a coherent story about your sources?

Estimated time per layer: Layer 1 (style) — 10–20 minutes. Layer 2 (consistency) — 15–30 minutes. Layer 3 (verification) — 30–60 minutes. Total: 45–90 minutes for a typical 30–50 reference manuscript.

⚠️ Common mistake: Assuming that because a reference manager auto-formatted your citations, they’re correct. Reference managers organize but don’t verify. A fabricated reference from ChatGPT will be formatted perfectly in APA style and still be completely wrong.

Tool Recommendations by Layer

Different tools excel at different layers. No single tool covers all three, which is why a layered approach is essential.

Free Tools

Paid / Freemium Tools

What we recommend: Use Zotero or EndNote for Layer 1 formatting, perform Layer 2 consistency checks manually (or with the Word plugin), and use Crossref Simple Text Query + Retraction Watch Database for Layer 3 verification. This free combination covers every audit layer without requiring paid tools.

Common Reference Errors That Trigger Desk Rejection

ReferenceChecker.org catalogued ten specific citation error types that trigger desk rejection or mandatory revision. Here are the most common:

1. Citation style inconsistency — Mixing APA and Vancouver within the same manuscript
2. Missing DOIs — References without DOIs when the journal requires them
3. Incorrect journal abbreviations — Using full journal names where abbreviated forms are required
4. Duplicate references — The same paper listed twice with different formatting
5. Citations without references — In-text citations that don’t appear in the reference list
6. References without citations — Reference list entries never cited in the body
7. Wrong volume/issue/page numbers — Metadata that looks correct but doesn’t match
8. Non-existent authors — AI-hallucinated author names (real names from adjacent fields)
9. AI-hallucinated citations — Completely fabricated papers (ghost references)
10. Unverified retracted papers — Citing papers that have been formally retracted

For context on retracted paper checking, see our guide on “How to Check if a Paper Has Been Retracted”.

What ICMJE Requires (The Gold Standard)

The International Committee of Medical Journal Editors (ICMJE) places verification responsibility squarely on authors:

“Authors are responsible for citing references accurately and should be able to attest that the references cited support the associated statement.”

This isn’t a suggestion. It’s the authoritative standard used by most medical and scientific journals as a baseline requirement. The ICMJE guidelines explicitly require authors to:

• Verify that every cited reference supports the claim being made
• Ensure reference accuracy (correct title, authors, journal, year)
• Use PubMed to verify biomedical references where applicable
• Check for retracted papers before final submission
• Use correct journal abbreviations from the NLM catalogue

If your target journal follows ICMJE guidelines (most do), this is the standard you need to meet. You cannot fulfill this requirement using a reference manager alone — it requires actively opening and reading each cited source to confirm it supports your claim.

The arXiv One-Strike Policy (Why This Matters Now)

arXiv introduced a one-strike ban in May 2026: authors whose submissions contain “incontrovertible evidence” they didn’t verify LLM-generated output — including hallucinated references — are banned from arXiv for one year. After the ban lifts, their papers must “clear peer review at a reputable venue” before arXiv will re-accept them.

This policy reflects a broader trend. Conference organizers at ICLR, NeurIPS, ICML, and ACL have implemented automated systems that desk-reject papers with confirmed hallucinated citations. Springer Nature and Elsevier have updated their editorial policies to explicitly treat AI-generated hallucinated citations as grounds for desk rejection or post-publication retraction.

For context on what journals specifically check, see our guide on “Do Journals Check References in 2026?”.

When to Run the Audit (Timing Matters)

Reference auditing should be integrated into your writing workflow, not saved for the last minute:

• During research: Resolve each DOI and verify the metadata as soon as you add a source. Don’t batch-verify later.
• During writing: As you draft, flag any references you haven’t fully verified. Keep a running list.
• Before submission: Run the full checklist above. This is where Crossref Simple Text Query is most useful.
• After submission (if required): Some journals require authors to verify sources. If so, you should have done this work already.

Quick First Check — What Takes 5 Minutes

If you only have five minutes before submission, do this:

1. Run the Crossref Simple Text Query on your reference list (paste one entry per line)
2. Check each DOI at doi.org — verify it resolves to the correct paper
3. Search the Retraction Watch Database for any titles flagged as retracted

These three steps catch approximately 80% of reference-level problems with minimal time investment. If any DOI fails resolution or any reference is flagged as retracted, fix it before submitting.

A pre-submission reference audit isn’t optional. It’s a five-to-ten-minute task that prevents desk rejection, revision requests, and post-publication corrections. The three-layer framework (style → consistency → verification) covers every error category that editors flag, and the checklist above gives you a concrete, copyable workflow you can run before every submission.

Run it once. Run it before every paper. It takes 45–90 minutes, saves weeks of revision cycles, and protects your reputation as a responsible researcher.