How to Check if a Reference Is Real: DOI, Crossref, and Database Verification Guide

Why Reference Verification Matters Right Now

Let’s get one thing straight: AI didn’t just make citation mistakes more common. It made them dramatically more common.

A 2026 audit of 2.5 million biomedical papers published in The Lancet found that fabricated citations appeared in 1 in 277 papers in the first seven weeks of that year. The rate in 2023 was 1 in 2,828 — meaning the problem grew 12 times faster in roughly three years.

Walters and Wilder (2023) found that 55% of citations generated by GPT-3.5 were entirely fabricated, while 18% of GPT-4’s were. The numbers for medical and mental-health papers are even higher.

This isn’t just about AI. Students copying citations from other papers’ reference lists without checking them, transcribing references from conference posters with typos, and citing papers they couldn’t actually read are all common human errors. AI just magnified existing problems.

The reason this matters so much is simple: when an examiner finds a fabricated citation, it doesn’t look like a mistake. It looks like academic misconduct. And the penalty — from failing a section to expulsion — is severe.

• A fabricated citation hit 1 in 277 papers in early 2026 — a 12-fold increase from 2023 — making reference verification more critical than ever.
• Five major databases provide overlapping but incomplete coverage. No single database is sufficient; you need at least two for reliable verification.
• A reference can have a perfectly valid DOI and still be wrong — the DOI may resolve to a completely different paper (the “Frankenstein citation” pattern).
• The three-step verification process (resolve DOI → search title in databases → read the source) catches all three types of problematic citations.

The Three Types of Problematic Citations

Before you learn how to check references, it helps to understand what you’re actually checking for. There are three distinct types of problematic citations, and each requires a different verification approach.

1. The fake paper (ghost citation)

The paper does not exist. The journal name is real, the authors sound plausible, the year is recent — but there’s nothing to find. This is the most common AI-generated citation pattern.

How to spot it: DOI resolution fails with a “DOI Not Found” error, and title search returns zero results.

2. The wrong paper (Frankenstein citation)

The paper exists, but it’s not the one you cited. A real DOI resolves to a completely different article. The title, author, and year in your citation are wrong. This pattern is called “Frankenstein” because AI assembles real components (a real author, a real journal, a valid DOI format) into a composite that doesn’t exist.

How to spot it: DOI resolves successfully, but the returned article title and authors don’t match your citation.

3. The misused source

The paper is real and correctly cited, but it doesn’t actually support the claim you’re making. This is the hardest to catch mechanically. You have to read the source.

How to spot it: You can only catch this by opening the paper and reading the relevant section.

Let’s break down each verification step.

How to Verify a Reference: Step by Step

Step 1: Resolve the DOI

If your reference includes a DOI (Digital Object Identifier), this is the single most important check you can perform. The DOI is a permanent identifier assigned to almost all peer-reviewed journal articles, and it always resolves to the official record for that paper.

What to do:

1. Copy the DOI from your reference (it starts with 10. followed by numbers, like 10.1038/nature14539)
2. Open your browser and type https://doi.org/ followed by the DOI in the address bar
3. Hit enter — you’ll be redirected to the publisher’s page for that article

What each outcome means:

• Resolves to a matching article: The DOI is valid and points to the correct paper. Great.
• Returns “DOI Not Found”: The DOI was never registered. This could mean the paper doesn’t exist, or the DOI was typed wrong.
• Resolves to a different article: This is the Frankenstein pattern — a real DOI pointing to the wrong paper.

After the DOI resolves, compare the returned article’s title, authors, and year with what’s in your citation. They must match exactly.

Pro tip: A retracted paper will still resolve at its DOI. The DOI tells you the paper exists; it doesn’t tell you whether it’s still valid. You need Step 2 and Step 3 for that.

Step 2: Search the Title in Multiple Databases

Even if the DOI resolves correctly, you should verify the citation against at least one independent database. Different databases use different indexing criteria and may catch errors that the DOI resolver misses.

Crossref

URL: search.crossref.org

Crossref is the single most important database for reference verification. It covers 90+ million registered DOI records across journal articles, books, conference papers, datasets, and more. Roughly 97% of peer-reviewed journal articles have a Crossref record.

How to use it:

• Enter the exact article title in quotes (e.g., "Title of the Paper")
• Add author names if you want to narrow results
• Look for the record that matches your citation
• Verify the authors, year, journal, volume, and page numbers match

What Crossref doesn’t cover: Pre-2000 articles (the DOI era started in 2000), books without DOI registration, dissertations, regional journals without English titles, and some non-English literature.

Google Scholar

URL: scholar.google.com

Google Scholar has the broadest coverage of any scholarly database — it indexes theses, preprints, books, patents, and court opinions. It’s useful as a catch-all when Crossref returns nothing.

How to use it:

• Search the exact article title in quotation marks
• Check the “[PDF]” link — if it appears, there’s an accessible version
• Look at the “Cited by” count — legitimate papers accumulate citations over time

Important limitation: Google Scholar has no metadata API and provides no structured search. A paper not found on Google Scholar doesn’t mean it doesn’t exist — it just means Google Scholar hasn’t indexed it. Never use Google Scholar as a “negative test” (i.e., don’t conclude “not found = fake”).

PubMed

URL: pubmed.ncbi.nlm.nih.gov

PubMed is the authoritative database for biomedical, clinical, and life sciences literature. If you’re working in a health-related field, this should be your first stop.

How to use it:

• Search the exact article title
• Check the PMID (PubMed ID) — if the PMID in your citation doesn’t match, something is wrong
• Look for “[Retraction]” or “[Publication Analysis]” tags in the record — these indicate the paper has been retracted or analyzed for issues

What PubMed doesn’t cover: Non-biomedical literature, non-English journals not indexed in MEDLINE, books, conference proceedings.

OpenAlex

URL: openalex.org

OpenAlex is the most comprehensive open bibliographic database available, covering 270+ million works. It’s built from Crossref, Semantic Scholar, PubMed/PMC, and other sources, and it includes works, concepts, institutions, and sources.

How to use it:

• Search by title, author, or DOI at the OpenAlex homepage
• Compare the returned record’s metadata (title, authors, date, source journal) with your citation
• It’s particularly useful for non-English works indexed by Semantic Scholar

WorldCat

URL: search.worldcat.org

WorldCat is the world’s largest bibliographic database, with 1.4 billion records from 14,000+ libraries. It’s essential for verifying books, book chapters, dissertations, and other non-article sources.

How to use it:

• Search by title, author, or ISBN
• Check if the record exists and matches your citation
• Particularly useful for verifying textbooks and monographs that won’t appear in Crossref or PubMed

Step 3: Read the Source

This is the step nobody wants to do, and it’s the one you can’t automate. After confirming a reference exists and the details match, you need to open the actual paper and verify that the claim you’re citing is actually supported by the source.

A paper can be real, correctly cited, and still be misused. This is called a “contextually inaccurate” citation — and Resnik and Hosseini (2026) classified it as a distinct error type below fabrication. A citation that is contextually inaccurate may constitute research misconduct under U.S. federal regulations when citations function as data (e.g., in review articles or bibliometric studies).

Verification Workflow: Which Database Should You Use When?

The database coverage matrix below shows which source types each database covers and which database to reach for when. Use this as your reference table when verifying citations.

The Database Coverage Matrix

When to use which database:

• DOI present? → Resolve at doi.org, then verify with Crossref or OpenAlex
• Biomedical topic? → Add PubMed search to Crossref check
• Book or chapter? → Search WorldCat after Crossref
• No DOI and Crossref returns nothing? → Try OpenAlex, then Google Scholar
• Need to check a whole reference list? → Use Crossref’s Simple Text Query tool

The key insight: Every reference should be checked in at least two databases. If Crossref finds a match, verify it with OpenAlex. If Crossref finds nothing, try OpenAlex and Google Scholar. If you’re working with books, add WorldCat.

What Each Verification Result Means

When you run a reference through your verification process, you’ll land on one of these outcomes:

Verified ✓

A database record matches the citation on title, authors, and year. The reference is real and correctly cited.

Metadata mismatch ⚠️

A record was found, but the cited year, title, or author differs from the database record. Both versions are shown in most verification tools. This is a very common result — most often caused by citing the online-first year instead of the print year, or by copying a citation from another paper’s reference list without checking the original.

Could not locate 🔍

No matching record was found in any database you checked. This does not prove fabrication. Many real sources — pre-2000 articles, books without DOI registration, regional journals, dissertations, government reports — are perfectly real and still missing from indexes. Treat “not found” as a to-verify item, not a verdict.

Retracted ❌

The matched record carries a retraction notice or expression of concern. A retracted paper can still be cited, but only knowingly and with the retraction acknowledged. If you didn’t know it was retracted, replace it with a non-retracted alternative.

Check failed ⏳

A lookup service was temporarily unavailable, so the result is inconclusive. Re-run the check later. This is rare but does happen — database outages occur.

Hallucinated Citation Statistics: How Bad Is It?

The numbers are sobering. Here’s a timeline of major studies:

The Lancet audit is the most comprehensive study ever conducted. It checked 2.5 million papers indexed in PubMed and found that approximately 4,046 contained fabricated references. The rate increased roughly 12-fold between 2023 and early 2026.

The pattern is clear: as AI-generated writing has become more widespread, the rate of fabricated citations has accelerated proportionally. But the problem isn’t confined to AI — human errors in citation formatting and verification happen constantly, and the consequences are equally severe.

Tips and Common Mistakes

Mistake #1: Assuming “no DOI” means “fake”

Many legitimate sources don’t have DOIs: books, dissertations, conference papers from before 2000, and grey literature from regional or specialized publishers. Always search the title in WorldCat or OpenAlex before concluding a reference is fabricated.

Mistake #2: Trusting formatting over verification

A fabricated citation looks exactly like a real one. Real journal name, plausible authors, correct DOI format — everything checks out on the page. This is why formatting is a poor test. You need database verification.

Mistake #3: Verifying the DOI but not the metadata

A valid DOI proves that a paper exists. It doesn’t prove that the paper you cited is the one that exists. Always compare the returned record’s title, authors, and year with your citation.

Mistake #4: Using only one database

Different databases cover different content. Crossref dominates journal articles. PubMed covers biomedicine. Google Scholar has the broadest reach but the least structure. WorldCat handles books. Using only one database leaves gaps.

Mistake #5: Assuming “not found” = “fabricated”

This is the most common and most costly mistake. A reference not found in Crossref could simply be a book without a DOI, a pre-2000 article, or a paper in a journal that doesn’t deposit metadata with Crossref. Try OpenAlex, Google Scholar, and WorldCat before concluding fabrication.

Free Bulk Verification: Crossref Simple Text Query

If you’re verifying an entire reference list (not just one or two citations), Crossref’s Simple Text Query tool is the most efficient free option.

How to use it:

1. Go to apps.crossref.org/SimpleTextQuery
2. Paste your reference list (one entry per line, no line breaks within a citation)
3. Crossref attempts to find exactly one DOI for each reference
4. Set “Include PubMed IDs” for biomedical references
5. Set “List all possible DOIs” for thoroughness

Limitations:

• Works best with alphabetical or numbered reference lists
• Doesn’t check for retracted papers
• Fuzzy matching means some references won’t be matched
• Maximum 1,000 references per submission

This tool is free and doesn’t require an account. Your reference list is processed directly from your browser — it’s not sent to a server.

When Should You Check References?

Reference checking should be part of your writing process from the start, not a last-minute panic session. Here’s a practical workflow:

During research: Every time you add a source to your bibliography, resolve its DOI and verify the metadata within that session. Don’t batch-check later.

During writing: As you draft, flag any references you haven’t fully verified. Keep a running list.

Before submission: Run a final verification pass on every reference in your bibliography. This is where Crossref Simple Text Query is most useful.

After submission (if required): If you’re submitting to a journal that asks you to verify sources, or if your supervisor or examiner flags a reference, you’ll need to have done this work already.

The Bigger Picture: Why Verification Is a Student Skill

Reference verification isn’t just about avoiding academic misconduct penalties. It’s one of the core skills of academic writing. Understanding where your sources come from, knowing which databases to trust, and being able to check whether a claim is actually supported by the literature — these are the skills that distinguish a strong researcher from a student who just stacks sources together.

The Lancet audit showed that even experienced researchers cite fabricated references, usually because they trust sources they got from AI tools without checking. The solution isn’t to avoid technology. It’s to develop verification habits that work alongside it.

FAQ

How do I check if a reference is real without a DOI?

Search the exact article title in Crossref, OpenAlex, Google Scholar, or WorldCat (for books). A real paper should appear in at least one database with matching authors and year. If no database returns a match, try searching by author name alone.

What is the difference between a fake citation and a Frankenstein citation?

A fake citation (ghost paper) is a reference that doesn’t exist at all — no paper, no journal, nothing. A Frankenstein citation combines real components: a real author, a real journal, a plausible title, and sometimes even a valid DOI format. Both look correct on the page, but neither can be verified against an authoritative database.

Can I use AI tools to check my references?

AI tools can help draft citations, but they cannot verify them. LLMs generate citations the same way they generate text: by predicting plausible patterns. They don’t have access to a database of real papers. Always verify AI-generated references manually using the three-step process above.

How do I check if a paper has been retracted?

Search the Retraction Watch Database (retractiondatabase.org) by title, or check the paper’s record on Crossref or PubMed for retraction notices. Retracted papers may still resolve at their DOI, so you can’t rely on DOI resolution alone to catch retractions.

Does a reference without a DOI mean it’s fake?

Not necessarily. Many legitimate sources don’t have DOIs: books, dissertations, conference proceedings, government reports, and pre-2000 articles. Always search these by title in WorldCat (for books) or OpenAlex before concluding fabrication.

References

• Topaz, M., Roguin, N., Gupta, P., Zhang, Z., & Peltonen, L.-M. (2026). Fabricated citations: An audit across 2.5 million biomedical papers. The Lancet, 407(10541), 1779–1781.
• Walters, W. H., & Wilder, E. I. (2023). Fabrication and errors in the bibliographic citations generated by ChatGPT. Scientific Reports, 13(1), Article 14045.
• Resnik, L. B., & Hosseini, J. D. (2026). Hallucinated citations produced by generative AI may constitute research misconduct. Accountability in Research.
• Linardon, J., Shah, S., & McHugh, R. (2025). Influence of topic familiarity and prompt specificity on citation fabrication by large language models. JMIR Mental Health.
• Chelli, M., et al. (2024). Hallucination rates and reference accuracy of ChatGPT and Bard for systematic reviews. Journal of Medical Internet Research, 26, e53164.
• Cabanac, M. (2026). The coming crisis of hallucinated citations in science. Nature.