Fast facts
- Only 26.5% of AI-generated references are correct — that means 73.5% contain errors or are entirely invented.
- A single DOI lookup at
doi.orgcatches the majority of fake references in under 10 seconds.- A working DOI does not prove quality: DOIs verify persistent identification, not peer review or journal reputation.
Most students never learned how to verify a reference. You were taught how to format them in APA or MLA — the commas, the italics, the hanging indent — but nobody explained how to confirm they actually exist. Then comes the moment when an AI-generated citation appears in your reference list, looks perfectly formatted, and you have no idea whether the paper behind it is real.
Here’s how to check.
Before diving into each tool, here is the master workflow. It maps every source type to the best verification method and the difficulty level you should expect.
| Source Type | Primary Tool | Secondary Check | Difficulty |
|---|---|---|---|
| Journal article (has DOI) | Resolve at doi.org |
CrossRef metadata comparison | Easy |
| Journal article (no DOI) | Google Scholar (title search) | Publisher website archive | Medium |
| Biomedical / clinical source | PubMed | CrossRef or PMID resolution | Easy |
| Book or monograph | WorldCat (ISBN or title) | Publisher catalogue | Medium |
| Conference paper | IEEE Xplore / ACM Digital Library | Conference proceedings website | Medium |
| AI-generated reference | doi.org + Google Scholar |
Publisher issue archive | Hard |
Bottom line: Start with the DOI. If the reference has a DOI, resolve it — that’s the fastest filter. If there’s no DOI, fall back to Google Scholar or the relevant database. Always do a secondary check for anything you plan to cite in a submitted paper.
The DOI (Digital Object Identifier) is your first and best defense against fake references. A DOI looks like this:
10.1037/a0018218
To verify it, go to https://doi.org/ and append the DOI at the end. For the example above: https://doi.org/10.1037/a0018218
Pro tip: Don’t trust the DOI alone. A valid DOI confirms the identifier exists, not that it belongs to the paper you’re citing. You still need to compare metadata — the title, authors, journal, and year.
A DOI is an identification system. It does not prove that:
A DOI means “this paper has a persistent identifier.” Nothing more. For quality assessment, you need separate tools like Beall’s Lists (a registry of suspect predatory journals) and journal ranking databases.
CrossRef is the official DOI registration agency — it’s not a third-party service or a startup tool. It is the primary authority for DOI metadata, managing 150+ million works from 20,000+ publishers.
Go to https://www.crossref.org/ and use their search. Enter the title or DOI and compare the result against your citation.
Match each field:
Mathematical inconsistency red flags from SourceChecker include:
If any of these mismatch, the reference is almost certainly fake.
When a source has no DOI (many older papers, conference proceedings, grey literature), use Google Scholar to search the exact title in quotation marks:
"title of paper here" site:scholar.google.com
Rule of thumb: Always cross-check against at least one additional database. If a paper appears in Google Scholar but nowhere else, search the publisher’s website. If it appears nowhere, the source is likely fabricated.
PubMed is the U.S. National Library of Medicine’s database. It contains 35+ million citations from biomedical and life sciences journals.
If you’re citing a medical, clinical, or health-related source:
PMIDs function like DOIs for PubMed-indexed articles. Every PubMed record includes an [AID] field showing the DOI or PMID. You can verify the article exists by confirming this field.
Books don’t use DOIs. Instead, WorldCat is the world’s largest library catalog, connecting millions of holdings across institutions.
When to use WorldCat: Any book or monograph in your reference list. If you can’t find the book in WorldCat or the publisher’s catalogue, it probably doesn’t exist — or you’ve misremembered the title.
Even a verified paper can be retracted. Retractions occur when a journal or publisher withdraws a paper due to errors, misconduct, or other issues. A retracted paper is still citable in some contexts (for discussing the retraction), but it should not be used as evidence for a claim.
Retraction Watch Database tracks 68,000+ retractions across all disciplines. Search by DOI, PMID, title, or author at https://retractionwatch.com/.
If you use Zotero, install the Retraction Watch integration. It automatically scans your library and documents for retracted works.
Free online tool at https://scholar-sidekick.com/. Check retraction status by DOI or PMID. Reports: “Retracted,” “Expression of Concern,” or “Correction/Erratum.”
Checking whether a quote actually appears on the page you cite is harder than verifying a reference. Here’s the process:
Unverified quotation repositories (like Quotations Page or similar sites) frequently misattribute quotes or detach them from real page numbers. Always verify against the primary source.
Frankenstein citations — also called chimera citations — are among the hardest AI hallucinations to detect because they combine:
A chimera citation looks legitimate at a glance. The DOI resolves. The journal exists. The authors exist. But the paper doesn’t.
Automated tools are useful when you have 30+ references to verify or are doing a final pre-submission pass. They cannot replace manual checking for content accuracy, but they catch the majority of formatting and DOI errors quickly.
| Tool | Best for | Free tier | Notes |
|---|---|---|---|
| CiteTrue (citetrue.com) | AI-powered citation verification | Yes | Batch checking |
| TrueCitation (truecitation.com) | Bulk reference list checker | Yes | Good for APA/MLA |
| CiteMe (citeme.app/tools) | APA, MLA, Harvard formats | Yes | Format conversion included |
| Research Gold (researchgold.org) | Crossref + PubMed cross-checking | Yes | Dual database |
| Citely (citely.ai) | CrossRef DOI + metadata matching | Yes | Queries CrossRef, flags mismatches |
| Litrevu (litrevu.com) | Compares Crossref, OpenAlex, Zotero | Yes | Aggregates 4 tools |
When to use automated tools: Large reference lists (30+), AI-assisted writing, final pre-submission verification, DOI format and resolution checking.
When manual checking is still needed: Content accuracy (does the source actually support your claim?), grey literature, retraction status, quotation page numbers, unusual source types.
If you’re using AI research tools (ChatGPT, Perplexity, etc.), verify every citation before you cite it. Don’t wait until the end. Checking as you go saves time because you fix errors in real-time rather than hunting for them in a completed reference list.
If someone else provided the references (a co-author, a group project partner, a professor), check everything in your reference list. It costs 10 minutes per paper and protects you from inadvertently citing fake or retracted work.
Relatable moment: You copy-paste ChatGPT’s reference list straight into your paper. You feel good because the formatting is perfect. Then you realize one DOI resolves to a completely unrelated paper about marine biology instead of your thesis on cognitive behavioral therapy. You could’ve saved the paper in 30 seconds.
| Mistake | Why it’s bad | How to fix it |
|---|---|---|
| Checking only the DOI | A valid DOI doesn’t prove the paper matches your citation | Resolve DOI → compare metadata field-by-field |
| Trusting citation managers | Zotero, EndNote, and Mendeley format citations but don’t verify them | Cross-check every reference in a database |
| Using only Google Scholar | Scholar misses paywalled papers and some publishers | Always cross-check against CrossRef or the publisher |
| Skipping retraction checks | A retracted paper is not evidence — it’s a liability | Retraction Watch or Zotero Retraction Watch plugin |
| Citing secondhand sources | If Source B cites Source A, verify Source A yourself | Always go to the primary source |
Verifying a reference doesn’t require advanced technology or expensive software. It requires a systematic approach:
doi.org — 10 seconds, catches the majority of fake citationsThe most important rule: A DOI confirms identity, not quality. Always compare the resolved metadata against your citation. A working DOI with garbled title, wrong year, or mismatched journal is still a fake reference.
If you have 30+ references or are working with AI-assisted writing, use automated tools like Citely or CiteTrue for a quick format and DOI pass. But always follow up with manual checking for content accuracy and retraction status.
No. A valid DOI confirms the identifier exists, but the paper it resolves to might not match your citation. Always compare metadata — title, authors, year, journal — against the DOI-resolved record.
Use Google Scholar to search the exact title. If that fails, check the publisher’s website or, for books, WorldCat. Absence from Google Scholar is not conclusive but is a useful secondary signal.
Check if the journal appears in Beall’s Lists (a registry of suspect predatory journals), verify it’s indexed in Scopus or Web of Science, and review the journal’s editorial board and publication fees. See our guide on journal legitimacy for detailed detection steps.
A retracted paper should not be used as evidence. It may be cited in a paper about the retraction, or to discuss why the finding was invalidated — but it should not support a claim. Always check Retraction Watch before citing.
Automated tools are great for format checking and DOI resolution across 30+ references. They cannot verify whether a source actually supports your claim, whether it’s retracted, or whether quotes are on the correct page. Always follow up automated checks with manual verification.
Need help verifying your references before submission? Contact our team for reference verification and citation audit services.