Frankenstein Citations: How to Spot Partially Incorrect AI References

• Frankenstein citations are real paper fragments assembled into fake references, typically a real author paired with a title they never wrote
• The NeurIPS 2025 audit found 100 hallucinated citations across 53 papers, with 27% being “partial attribute corruption”
• Detection depends on the failure type: chimera citations need exact-title searches, distorted citations require DOI cross-checks, and compound deception demands claim verification
• Watson (2024) found most student submissions with hallucinated citations were never verified before submission
• Every verified AI-generated citation needs a five-step workflow, not a single click

You cited a reference your AI gave you. You haven’t checked it.

Here’s the thing about AI-generated citations: they look exactly as plausible as real ones. The formatting is spot-on. The journal name sounds right. The author has real publications in that field. But the specific paper that you put in your bibliography was never written.

This isn’t a new problem. It’s been happening since AI writing tools got popular. But it’s gotten worse. A 2026 audit of papers accepted at NeurIPS (one of the top AI conferences in the world) found 100 hallucinated citations across 53 accepted papers. Expert reviewers missed every single one.

The most dangerous of these hallucinations aren’t the ones that are completely invented. They’re the ones that look correct at a glance, with real author names, real journal titles, and real DOIs, but stitch them together into a paper that never existed. This is what researchers call Frankenstein citations.

This isn’t a hypothetical concern. A Nature investigation in 2026 estimated that tens of thousands of publications from that year contained AI-polluted references generated by language models. The problem isn’t limited to undergraduate papers — it has surfaced in peer-reviewed conference proceedings, journal articles, and professional consulting reports.

What Are Frankenstein Citations?

Frankenstein citations are AI-generated references that assemble fragments of real scholarly work into a citation that looks complete but describes a paper that never existed.

The term was coined by Joe Shockman, co-founder of Grounded AI, in analysis published by Nature. He likened the citations to the fictional Frankenstein monster: assembled from pieces of real things, but recombined into something that doesn’t actually exist.

This isn’t a hypothetical concern. A Nature investigation in 2026 estimated that tens of thousands of publications from that year contained AI-polluted references generated by language models. The problem isn’t limited to undergraduate papers — it has surfaced in peer-reviewed conference proceedings, journal articles, and professional consulting reports.

Here’s exactly what a Frankenstein citation looks like:

• The author name is a real researcher with real publications
• The journal title is a real, legitimate academic journal
• The paper title sounds exactly like something they would have written
• The DOI, year, and volume are formatted correctly

But the actual paper does not exist. No paper with that title was ever published in that journal by that author. The AI simply predicted the most plausible-looking combination of bibliographic fragments.

The “Uncanny Valley” Tell

The most reliable indicator of a Frankenstein citation is what researchers call the “uncanny valley” tell: the citation is structurally perfect, all the fields are present, all the formatting is correct, but the individual pieces have never actually belonged together. Every single element checks out when you look at it separately. Only when you verify the combination do you discover the paper was invented.

A quick-author search confirms the researcher exists. A journal name search confirms the journal exists. But a search for the exact title returns nothing. The citation is a plausible Frankenstein of authentic limbs, a chimera that survives casual checks but fails verification.

How It’s Different from “Vibe Citing”

GPTZero introduced the term “vibe citing” to describe the same phenomenon from the LLM’s perspective: citations that sound plausible enough to pass a reviewer’s quick glance. While “Frankenstein citation” describes the structure of the reference itself (real parts, fake combination), “vibe citing” describes the mechanism (the model predicts a plausible sequence and the reviewer trusts the vibe rather than verifying).

Both terms refer to the same core problem: AI-generated references that look real but aren’t, and the failure of human verification to catch them.

How Frankenstein Citations Are Built

An LLM doesn’t look up a database when it generates a citation. It predicts the next plausible string of text from patterns it has seen in millions of published references. When you ask for “three peer-reviewed studies on X,” the model produces:

A name from that field + a journal that publishes in that field + a plausible title + a DOI in the right format.

Every individual token is statistically probable. The combined reference reads perfectly. But the underlying paper does not exist.

The Three Types of Partial Corruption

Not all hallucinated citations fail the same way. Understanding the failure type determines how you catch it.

TextSight.ai’s taxonomy of AI citations identifies three common shapes. The NeurIPS 2025 audit and the Ansari et al. arXiv paper (2602.05930) classify them by their primary failure mode.

Type 1: Chimera (The “Frankenstein” Citation)

A chimera reference is assembled from real parts that never belonged together:

• A genuine author name
• A real journal
• An actual paper title

But the combination, that specific author writing that specific paper in that specific journal, never existed.

Example:

Tversky, A., & Kahneman, D. (1979). Heuristic anchoring under conditions of statistical uncertainty. Cognitive Psychology, 11(2), 168-192.

Tversky and Kahneman are real, foundational researchers. Cognitive Psychology is a real journal. 1979 is inside their active collaboration. Every part checks out individually.

But that exact paper with that exact title was never published.

How to catch it:

• Search the exact title in quotation marks on Google Scholar
• Check the author’s own publication list (ORCID or institutional profile)
• Verify the DOI resolves to that specific paper

This is the trickiest of the three because every piece passes a quick individual check. You have to verify the combination, not the parts.

Type 2: Distorted (The Real Paper, Wrong Details)

A distorted citation is a real paper underneath, but with details mangled:

• The year is off by two
• Two DOI digits are transposed
• The journal name is close but not exact
• The DOI points to a different article

These are dangerous because the source exists, so a casual “does this paper exist?” check passes, while the claim you attribute to it may not be supported by the actual text.

Example:

Bandura, A. (1977). Self-efficacy: Towards a unifying theory of behavioral change. Psychological Review, 84(2), 191-215.

Real. Famous. Correct.

A distorted version might read:

Bandura, A. (1978). Self-efficacy: Towards a unifying theory of behavioral change. Psychological Review, 84(3), 191-215.

Year is off by one. Issue number shifted. This is the kind of distortion most students miss entirely.

How to catch it:

• Resolve the DOI at doi.org
• Compare every field against the resolved page: title, authors, year, volume, issue, page numbers
• If any field disagrees, the citation is distorted

Type 3: Compound Deception (The Layered Lie)

Compound deception is when multiple failure modes stack on top of each other. The Ansari et al. NeurIPS paper found that every single hallucinated citation in their analysis exhibited compound failure modes; no hallucination was a simple, single-mode error.

The distribution of secondary characteristics was dominated by:

• Semantic hallucination (63%) — the paper is real, but the claim you attribute to it isn’t there
• Identifier hijacking (29%) the DOI resolves but to a completely unrelated paper

What this looks like in practice:

An AI tool gives you a citation that is mostly real: the author, the title, the DOI are all correct, but the claim you make about what the paper “found” or “concluded” is fabricated. This is the most insidious type because the reference itself looks perfect, and nobody reads every source they cite.

How to catch it:

• Resolve the DOI. It should point to a real paper
• Read the relevant section and confirm it says what you’re citing it for
• If the claim isn’t supported, it’s compound deception

How to Detect Each Type

Detection depends on the failure mode. You can’t use the same tool for all three.

For Chimera Citations

The tool: Google Scholar (exact-title search)

The test:

1. Copy the exact paper title into quotation marks
2. Paste it into Google Scholar
3. If an exact hit appears, that’s good
4. If close-but-not-quite hits appear, watch out
5. If nothing appears, it’s fabrication

Why this works: A real, published, indexed paper with a specific title almost always shows up in Google Scholar. Silence is the strongest signal of an invented reference.

What happens in practice is straightforward. Copy the entire title into quotation marks. Google Scholar will search for exact matches. If the paper was actually published, Google Scholar will return it as the first result. If the AI made up the title, the search will return nothing or unrelated papers. This one action catches the majority of Frankenstein citations in a single pass.

For even stronger verification, check the author’s own publication list. Most researchers list every paper they’ve written on their ORCID profile, Google Scholar page, or university directory. If the paper you’re citing isn’t on that list, it probably wasn’t written by that author.

For Distorted Citations

The tool: doi.org resolution + CrossRef metadata

The test:

1. Paste the DOI after https://doi.org/ in your browser
2. Read the landing page
3. Compare the title, authors, year, journal, and page numbers against your citation
4. Every field must match

Why this works: The DOI is the hardest part of a citation for an AI to fake convincingly and the easiest part for you to test. If the DOI resolves to a different paper, you’ve caught a distorted or chimera reference.

The https://doi.org/ resolver redirects you directly to the publisher’s page for that exact article. What happens next matters: read the title, the authors, and the publication year on the landing page. If these don’t match your citation, you’ve found a distortion or chimera.

Pro tip: Don’t just confirm that the DOI resolves. Read the landing page. A DOI can resolve to a real paper, and the real paper is completely unrelated to the citation you were given. This is the “Topaz Pattern” documented in the Lancet audit: a real DOI paired with a fabricated title.

In practice, here’s what the process looks like. Copy the DOI from your citation. Paste it into https://doi.org/[DOI-number] in your browser. The resolver page will show you the article’s official metadata — title, authors, publication date, journal, volume, issue, and page range. Compare every single field against what you wrote down. Any discrepancy means the citation is wrong.

For Compound Deception

The tool: Reading the actual source

The test:

1. Confirm the paper exists (DOI resolves)
2. Open the source
3. Read the section you’re citing
4. Confirm it says what you’re attributing to it

Why this works: Even when the paper is real, the claim you cite may not be supported by the actual text. This is the step people skip, and it’s the one that catches semantic hallucinations.

This is where most students cut corners. They resolve the DOI, confirm the paper exists, and then proceed to quote it without reading the relevant section. They assume the AI described the paper accurately. That assumption is often wrong.

The way to catch compound deception is simple but time-consuming: read the section you’re citing. Don’t skim. Don’t rely on the abstract. Go to the actual paragraph, read the surrounding context, and confirm the specific claim is there. If the paper is about topic X and you’re citing it for topic Y, you’ve caught a Frankenstein citation — the paper exists, but the citation misuses it.

The Frankenstein Detection Checklist

This is the primary tool for verifying any AI-generated citation. Run it for every reference before you submit.

Bottom line: If any step fails, the citation is fabricated, distorted, or compound-deceptive. Do not cite it.

The Frankenstein Anatomy: Real vs. Fake Breakdown

The TextSight.ai framework (“Nobody Wrote That Paper”) breaks AI citations into three categories, and each category has its own verification pattern. Use this table to quickly diagnose which type of Frankenstein citation you’re dealing with.

This table shows why Frankenstein citations are so dangerous: every individual fragment passes a quick check. Only when you verify the combination of the specific author writing the specific paper in the specific journal with those specific details do you discover the reference was invented.

Citation Tools: Which One Catches What?

Manual verification catches everything. Automated tools are getting better but still can’t replace checking every citation yourself.

Here’s a brief comparison of the main detection tools:

What we recommend:

• For most students: Use the five-step manual checklist above. It’s free, it works, and it takes about two minutes per citation.
• For heavy reference lists (50+ citations): Use Hallucinator or Recite as a first-pass filter, then manually verify every flagged reference.
• For research papers: Cross-check your bibliography against Zotero’s best practices guide, then verify every citation.

For more on managing verified citations at scale, see our Citation Management Tools guide.

What to Do When You Find a Frankenstein Citation

So you’ve run the checklist, and one of your AI-generated citations failed. Maybe the title returned zero results. Maybe the DOI resolved to a completely different paper. Now what?

Here’s the fix: replace the fake citation with a real one that supports the same claim.

Don’t try to “fix” the DOI or edit the year. If the paper doesn’t exist, you need to find a real source for the point you were trying to make.

Use these search strategies:

• Google Scholar: Search the key concepts of your claim, not the fake title. Sort by relevance, scan the top results, and read the abstracts to confirm a paper genuinely supports your point.
• Subject databases: PubMed for medicine and life sciences, OpenAlex and Semantic Scholar for broad coverage, plus discipline-specific indexes (PsycINFO, JSTOR, IEEE Xplore, ERIC).
• Your library: University libraries give you licensed access to paywalled databases and, crucially, a research librarian whose job is helping you find the right source fast.

Read enough of each candidate to confirm it actually says what you’re claiming. This is the scholarship the AI skipped, and the reason the final work ends up yours instead of a model’s guess.

Once you’ve replaced the citation, check the rest of your references. Fake citations tend to cluster because the conditions that produced one were in force for the whole session: the model guessing instead of retrieving.

What Students Should Know

A 2024 study by Alex P. Watson at the University of Mississippi examined AI-generated citations submitted by freshmen. The findings are important for understanding the scope of the problem:

• Most suspect citations were not random nonsense: they mixed real fragments (author names, journal titles, publication years) with invented ones (volume numbers, page ranges, article titles)
• The fabricated references retained plausible elements of real scholarship, making them easy to trick casual or superficial checks
• Most student submissions with hallucinated citations were never verified before submission
• Submitting fabricated sources, even unintentionally, frequently triggers academic integrity reviews

The study is published in The Serials Librarian (Vol. 85, Issue 5-6, 2024, pages 172-180). The full article is available at https://www.tandfonline.com/doi/full/10.1080/00404055.2024.834923.

Resnik & Hosseini (2026) argue in Accountability in Research that hallucinated citations may constitute research misconduct when citations function as data in scholarly papers. Their analysis shows that when a researcher uses a GenAI tool without checking its output for veracity, they are acting recklessly because they’ve shown indifference to the risk of data fabrication.

“To act recklessly means to propose, perform, or review research, or report research results, with indifference to a known risk of fabrication.” — Resnik and Hosseini, 2026

The key takeaway for students: “The AI gave it to me” is not a defense. Every university treats citation accuracy as your responsibility, not the AI’s. Verifying what you cite is not optional.

Conclusion: Verification Is a Habit, Not a Choice

Frankenstein citations aren’t a quirk of AI writing tools. They’re a predictable byproduct of how large language models generate text: they predict the next plausible sequence, not verify whether the source exists. The result is a citation that looks perfect to anyone who glances at it: real author name, real journal title, correct formatting, but describes a paper that was never written.

Catching them requires a single, simple habit: verify every AI-generated citation before you submit it. Not after. Not halfway through the bibliography. Every citation, one by one, using the five-step checklist above.

It takes about two minutes per citation. That’s less than ten minutes for a paper with five references. The cost of verification is trivial, while the cost of citing a fabricated source ranges from a lowered grade to an academic integrity hearing.

Don’t trust the AI to verify its own work. Don’t trust a “working” DOI to prove a paper is real. Don’t trust the vibe. Run the checklist, catch the Frankenstein, and submit something you can defend.

FAQ

What exactly is a Frankenstein citation?

A Frankenstein citation is a real paper fragment assembled into a fake reference, typically a genuine author name paired with a real journal title and a paper that was never written. The term was coined by Joe Shockman of Grounded AI in Nature‘s analysis of AI-polluted literature. Unlike completely invented references, Frankenstein citations survive casual human checks because every individual piece looks real.

How do I detect Frankenstein citations?

Use the five-step checklist above: (1) exact-title search in Google Scholar, (2) DOI resolution at doi.org, (3) cross-reference metadata with CrossRef or PubMed, (4) author verification via ORCID or institutional listing, (5) claim alignment by reading the actual source. If any step fails, the citation is fabricated.

Are Frankenstein citations caught by DOI resolution alone?

No. This is the biggest reason Frankenstein citations are so dangerous. A DOI can resolve to a real paper, but to a different paper, so the model copied a real DOI and attached it to a fabricated title and author. Only after resolving the DOI and comparing the landing page against your citation do you discover the mismatch.

What is “vibe citing”?

GPTZero coined “vibe citing” to describe citations that sound plausible enough to survive peer review without verification. While “Frankenstein citation” describes the citation structure (real parts, fake combination), “vibe citing” describes the LLM’s mechanism (predicting plausible text). Both refer to the same problem.

Can an AI citation checker catch Frankenstein citations?

Automated tools like Recite, Hallucinator, and GPTZero’s hallucination checker can catch many types of fabricated citations, especially fully invented ones. However, they can miss compound deception (semantic mismatches where the paper is real but the claim isn’t). Manual verification remains essential.

Is citing a Frankenstein citation academic misconduct?

Most universities treat citing fabricated sources, even unintentionally, as a serious matter, regardless of intent. Watson (2024) found that submitting fabricated sources triggers academic integrity reviews at most institutions. The accuracy of your references is your responsibility, not the AI’s.

How often do AI tools generate fabricated citations?

Studies consistently show that a meaningful share of AI-generated citations are invented. Nature’s Scientific Reports documented substantial fabrication rates in ChatGPT’s bibliographic output, with estimates commonly ranging from 15-30% fully fabricated citations and another 20-30% containing real-looking but incorrect details. The exact rate varies by model, field, and topic familiarity.

Related Guides

• Fake LLM Citations: Why AI Fabricates References: The mechanism behind AI citation hallucinations and statistics on fabrication rates
• How to Verify a Link (DOI/Crossref/PubMed) Is Real, Not Fake: The four-step verification workflow that catches every fabrication pattern
• How to Cite AI Chat Conversations: APA, MLA, Chicago Guide 2026: Proper citation formats for AI-generated content
• Citation Management Tools: Zotero vs Mendeley vs EndNote 2026: Tools for organizing and verifying citations