A bold challenge has been taken up by a determined reader, and the results are eye-opening!
This summer, we covered a book with fake citations, and our readers responded with a flood of similar examples. One reader, Michał Wójcik, a Ph.D. student, decided to take matters into his own hands and create a program to uncover these problematic texts.
The issue of fake references in academic books is a serious one, and Wójcik's work sheds light on a potential crisis in the publishing world.
His Python script searches through books, verifying citations by checking DOIs on Crossref. He manually checks the tricky ones, spending hours on each book.
Wójcik focused on books published in 2025 and checked 22 with varying certainty. The two he sent to us were particularly concerning.
The first, an urban planning book, contains citations with substantial errors or references to non-existent works. The second, on energy storage, was translated into English by AI, including all the citations. Both were published by Springer Nature.
Large language models like ChatGPT often generate fake citations, and publisher guidelines often forbid wholesale AI-generated text. Yet, AI is used in other publishing aspects, like translation, with varying disclosure requirements.
Springer Nature has retracted chapters and entire books for unverifiable references. Wójcik's script identified 40 missing DOIs in the urban planning book, and our own checks found 11 unverifiable citations.
But here's where it gets controversial...
We contacted the listed authors of these works, and some confirmed they did not write them or that there were significant errors. One researcher, Bogdan Ibanescu, said, "I am not sure if this is a fake citation of mine or another author."
Manolya Kavakli-Thorne, a professor, confirmed she did not write an article cited with her name, and believes it's an example of deep learning hallucinations.
Other citations had unusual errors, like extra authors added to an article written by one person.
When asked about AI involvement, the author of the urban planning book, Gouri Sankar Bhunia, said they used an LLM tool like ChatGPT for minor purposes like checking grammar and citation formats.
Springer Nature's AI policies permit such use, but the company emphasizes human accountability and author agreement.
Bhunia assured us that all citations were prepared and verified by academic professionals, and any errors were unintentional.
Springer Nature was already investigating both books when we contacted them, and they plan to issue corrections and retractions.
The second book, translated from Chinese, primarily references Chinese-language sources, but many citations lack DOIs or functioning hyperlinks.
Wójcik believes scientific references should allow readers to access the resources, which is often not possible with translated titles and missing links.
And this is the part most people miss...
Wójcik has expanded his script to detect sham references in papers too, checking around 100,000 so far. His script flags about 1 in every 300 papers, and most have nonsensical references.
He believes such problems should not exist in scientific literature, especially with online versions providing links to Google Scholar.
Springer Nature's spokesperson acknowledges that errors in book citations are rare but experienced by all publishers, typically identified through editorial review, peer feedback, and post-publication scrutiny.
Wójcik's work is a reminder of the importance of fact-checking and the potential pitfalls of AI-assisted writing and translation.
What are your thoughts on this issue? Do you think AI-generated content should be more strictly regulated in academic publishing? We'd love to hear your opinions in the comments!
