AI Fake Citations in a Peer-Reviewed Z-Library Paper: What Went Wrong

AI fake citations in a peer-reviewed Z-Library paper

🇻🇳 Đọc bản tiếng Việt

AI fake citations have turned up in a peer-reviewed paper about Z-Library, published this month in the journal New Review of Academic Librarianship by Taylor & Francis. TorrentFreak, the news site the paper cites, says it never wrote the article attributed to it, and it could not locate at least 10 of the paper’s 44 references. The authors’ acknowledgements say they used Consensus AI and ChatGPT during research and writing.

AI fake citations in a peer-reviewed Z-Library paper

Key Takeaways

  • A peer-reviewed paper on Z-Library, written by two librarians at Nigerian universities, was published earlier this month in New Review of Academic Librarianship (Taylor & Francis).
  • It cites a 2021 TorrentFreak article, “Publishers continue crackdown on shadow libraries”, that TorrentFreak says it never published.
  • TorrentFreak had trouble finding at least 10 of the 44 references, and others did not back the claims they were attached to.
  • The acknowledgements say Consensus AI helped sift literature and ChatGPT improved the language of parts of the paper.
  • The AI detector Pangram flagged the abstract and conclusion as “100% of the text is likely AI-generated”. Such tools can produce false positives.
  • Taylor & Francis has referred the case to its Editorial and Publishing Ethics & Integrity teams.

Watch: our 30-second video summary (in Vietnamese).

What Is the Z-Library Paper About?

Z-Library is one of the largest pirate ebook libraries online, offering millions of books and papers for free. Publishers have fought it with takedown notices and blocking orders, and U.S. authorities indicted its alleged operators.

The paper, titled “Z-Library and Full-Text Access: Democratizing Knowledge or Undermining Copyright Ethics?”, looks at this tension. It does not pick a side. Its conclusions are not especially newsworthy. The problem is in the reference list, where AI fake citations appear to sit next to real ones.

The AI Fake Citations: Full List of Sources That Could Not Be Found

A citation is a reference to an earlier work that supports a claim. A “fake” or “hallucinated” citation looks real, with a plausible title, a real journal and a real expert’s name, but the work does not exist or says something else. Here is what TorrentFreak reported:

Cited sourceWhat the paper claimsWhat TorrentFreak found
TorrentFreak, 2021, “Publishers continue crackdown on shadow libraries”Elsevier and Springer Nature took legal action similar to the Z-Library enforcementTorrentFreak never published an article with that title; the link points to its homepage
U.S. Department of Justice press releaseThe DOJ “arrested individuals” in 2022The DOJ said the two alleged operators were arrested in Argentina “at the request of the United States”; the press release title given in the paper does not exist
Daniel Gervais, 2019, Harvard Journal of Law & Technology, “The future of intellectual property in the shadow of shadow libraries”Supports the paper’s legal discussionThe issue’s table of contents does not include it; the cited pages hold two articles on other topics
Carol Tenopir and colleagues, 2019, College & Research LibrariesSupports claims about library useThe cited issue has no article by Tenopir; she co-wrote a 2019 article on a similar topic in a different journal, under a different title
Björn Brembs, 2019Supports the paper’s argumentsCould not be located
Articles credited to Smith, Piper, Okike and AronsonVarious supporting claimsCould not be located
Kaspersky report on “the hidden dangers of shadow libraries”Security risks of shadow librariesCould not be located

In total, at least 10 of the 44 references could not be found. TorrentFreak notes that what Elsevier and Springer Nature actually did was obtain a UK blocking order against Sci-Hub, a different site. Several Z-Library domains were also temporarily suspended over copyright claims. But it was not aware of the specific action the paper describes.

TorrentFreak asked Gervais, Brembs and Tenopir whether they wrote the papers credited to them. None of them responded. The lead author shared a copy of the paper but did not answer follow-up questions about the references.

Did AI Create These Fake Citations?

The evidence points toward AI, but it is not proof. Three facts stand out:

  1. The authors disclosed AI use. “We confirm that Consensus AI was used in sifting through large volume of literature and extract relevant information,” the acknowledgements say. ChatGPT was used to improve the language of parts of the paper.
  2. The pattern matches known AI behaviour. AI chatbots are known to invent references, often with plausible titles credited to real experts in the field. That is exactly what these AI fake citations look like.
  3. An AI detector flagged the text. Pangram reported that the abstract and conclusion were “100% of the text is likely AI-generated”. These tools are not perfect and can give false positives.

TorrentFreak is careful here: the findings could be explained by AI, but that does not prove AI wrote the references.

How Did Taylor & Francis Respond?

Taylor & Francis replied to TorrentFreak and confirmed it is looking into the matter. “I can confirm that this has been referred to our Editorial and Publishing Ethics & Integrity teams, who will investigate the concerns you have raised,” a spokesperson said.

The publisher will not comment further while the investigation is ongoing. Whether the paper will be corrected or retracted is still unknown.

There is also an irony TorrentFreak points out. The paper warns that Z-Library hurts publishers’ revenue, which funds peer review and editing. It does not mention that peer reviewers generally are not paid, an issue central to a recent lawsuit against major academic publishers, including Taylor & Francis. And many rightsholders are suing AI companies for training models on shadow libraries, while this paper about a shadow library may itself contain AI fake citations.

Why AI Fake Citations Matter for Everyday AI Users

This case shows that AI fake citations can pass through peer review, a process meant to catch errors before publication. If they can fool reviewers at a commercial journal, they can easily fool a student, a blogger or a marketer.

The risk is the same whether you use ChatGPT, Consensus AI or another research tool. A reference that looks perfect can still be invented. We have covered similar reliability questions in our piece on using the Lean theorem prover for AI reliability and in our Perplexity AI review, which looks at how an AI search tool shows its sources.

How to Check AI Citations Before You Use Them

You do not need special software to avoid AI fake citations. These steps follow directly from how TorrentFreak caught the problems:

  1. Open every link. The fake TorrentFreak citation linked to the site’s homepage, not to an article. A citation that points to a homepage is a warning sign.
  2. Check the journal issue. TorrentFreak opened the table of contents of the cited issues. If the article is not listed, or the page numbers belong to other papers, the citation is wrong.
  3. Search the exact title. If a title cannot be found anywhere, treat it as invented until proven otherwise.
  4. Match claims to the source. Some references in the paper existed but did not back the stated claims. Read the part you are citing.
  5. Go to the primary source. The DOJ press release said the arrests happened in Argentina, a detail the paper got wrong.

Privacy matters too when you paste research into chatbots; see our guide to AI chatbot privacy trackers.

Is AI Still Useful for Research Despite AI Fake Citations?

Yes, with limits. Tools like Consensus AI can help sift a large volume of literature, and ChatGPT can polish language. The weak point is trusting the output without checking. AI fake citations are dangerous precisely because they look plausible, mix real names with invented titles, and sit next to genuine references.

Pros: faster literature screening, better language for non-native writers.
Cons: invented references, claims that do not match sources, and the risk of retraction or reputational damage.

AI Fake Citations FAQ

What are AI fake citations?
They are references generated by an AI tool that look real, often with plausible titles and real experts’ names, but point to works that do not exist or do not support the claim.

How many references in the Z-Library paper could not be found?
TorrentFreak had trouble finding at least 10 of the paper’s 44 references, and others did not necessarily back the stated claims.

Which AI tools did the authors use?
The acknowledgements say Consensus AI was used to sift literature and extract information, and ChatGPT was used to improve the language of parts of the paper.

Has the paper been retracted?
Not yet. Taylor & Francis has referred the concerns to its Editorial and Publishing Ethics & Integrity teams. Whether the paper will be corrected or retracted has yet to be seen.

How can I spot AI fake citations?
Open each link, check that the article appears in the cited journal issue, search the exact title, and confirm the source actually says what is claimed.

Sources

Last updated: October 24, 2026

Trạm Công Nghệ AI
Logo
Compare items
  • Total (0)
Compare
0