The Journal Editor Asked to Produce a Paper She Never Wrote

An adjunct education professor wanted to read a paper. Standard stuff, the kind of favor academics ask each other constantly: hey, can you send me a copy of your 2023 piece on instructional presence in eLearning? She emailed the two credited authors, Charles Hodges and Stephanie Moore, and asked for the reprint.

They went looking for it. There was nothing to find. The paper "Instructional presence and learner success in synchronous and asynchronous eLearning" never existed. Yet someone's works-cited list said Hodges and Moore wrote it in 2023.

This would be a strange enough story if Hodges and Moore were random names. They carry weight. Moore is an organization, information and learning sciences professor at the University of New Mexico, and editor in chief of the Journal of Computing in Higher Education, a Springer title. Hodges is a professor at Georgia Southern University who spent a decade, 2014 to 2024, as editor of TechTrends. Between them they have reviewed and rejected more citations than either could count. Neither one could produce the paper apparently sitting under their own names.

The citation was dressed for the occasion: proper APA formatting, a real host journal, the Online Learning Journal, and a DOI link that looked exactly like every other DOI link you've ever clicked without thinking about it. Click this one and you land on a page that says, politely, DOI not found. A ghost with a working shirt and tie, no body underneath.

It happened again. Within the last two months, Hodges was asked to review a book proposal that listed, among the competing titles on the market, a book supposedly edited by Hodges and Moore, published by Springer. Also invented. Also confident enough in its own existence that it made it into a proposal meant to survey the real competition.

Hodges and Moore are the visible tip. York St. John University geography professor Pauline Couper reviewed a grant application that cited a nonexistent paper apparently by her, and said as much, flatly, on Bluesky: "cited a nonexistent paper, apparently by me." Gale Sinatra, an education and psychology professor at USC, did the natural thing out of curiosity and asked an AI chatbot to list her own publications. It handed back a mix of real papers and invented ones, sitting side by side with no visible seam between them.

The mechanism

This is the machine working as designed. Nobody set out to lie. A chatbot asked for a citation predicts the next plausible sequence of words, author names that sound right together, a journal that plausibly runs that kind of paper, a year that fits the argument, a DOI formatted the way DOIs are formatted. It builds the citation the way it builds every other sentence: by pattern. Facts never enter it. Whether the paper is real sits entirely outside what the model is doing. It will hand you the invented reference with the same fluent confidence it uses for the real ones, because to the model there is no difference. A citation is just a shape it has seen ten million times, reproduced on demand.

That is how you end up with a DOI resolving to nothing, a book that was never printed, a paper that two working academics cannot find under their own names. The tool is guessing. It never tells you which parts were guesses.

The editors are already exhausted

Andrea Harkins-Brown edits the Journal of Technology and Teacher Education. She's been in the chair about a year and already sees more AI-fabricated citations in submissions than when she started. The scary part is when the fakes get caught. "We've had papers that have been through several rounds of review, I've looked at them as the editor and several rounds of reviewers have looked at them. And these phantom references are really hard to spot, we didn't catch them until we were in the copyediting phase." Multiple qualified reviewers over multiple passes, and the fake reference rides along undetected until someone is literally fixing commas.

When Harkins-Brown does catch one, the process is blunt: ask the author to send a copy of the cited paper. Most of the time, they can't. Springer runs its own integrity screening on submissions to Moore's journal and sometimes pulls papers before they even reach her desk, a filter that catches some of this, which is exactly how we know it doesn't catch all of it.

Peer review used to run on a baseline of trust: a citation existed unless someone had a specific reason to doubt it. That baseline is gone. The new default is that a citation is fabricated until the author proves otherwise, which is a genuinely strange thing to have to say about a scholarly footnote.

The paper about whistleblowing that needed its own whistleblower

In April, the Journal of Academic Ethics, a Springer Nature title, published a paper on the whistleblowing experiences of individuals with disabilities in Ethiopian public educational institutions. Erja Moore, an independent researcher in Finland who had previously presented on fake references turning up in Finnish master's theses, found it while researching a whistleblowing case of her own. She did what nobody upstream apparently had: she checked all 29 references.

At least 19 of them were fabricated by her count. Eighteen of the Google Scholar links sitting in the paper's own online reference section led nowhere, the digital equivalent of a return address that doesn't exist. Eleven of the fake references pointed to the Journal of Business Ethics, another Springer Nature journal, which means the fabrications weren't even reaching outside the family. A paper about integrity, invoking a sister journal's authority, using citations to that journal that the journal never published.

The corresponding author, Yelkal Mulualem Walle of the University of Gondar's Department of Information Technology, told Retraction Watch plainly that he and his coauthors used ChatGPT to generate the references. Then came the line that does all the work: "the research is real and used real data." A whistleblowing study, built on genuine fieldwork, wrapped in a bibliography made of nothing. The house was real. The foundation was drawn on by hand and labeled concrete. Two coauthors, from the University of Gondar and Ethiopia's Federal Ethics and Anti-Corruption Commission, did not respond to requests for comment, which is its own kind of answer.

Springer Nature's Michael Stacey said the publisher takes such concerns "extremely seriously" and is "looking into the matter carefully." Chris Graf, the company's research integrity director, was more candid about why this is hard to fix at scale: "This is more complex than it may at first appear, as references can be detailed by authors in a variety of different ways, often do not include DOIs, and simple tools to identify hallucinated references can produce false positives." Translation: you cannot just run a script over a bibliography and trust the output. Somebody has to actually go check whether the paper is there.

Guilty until proven cited

That's the state of the field now. An editor with a decade of experience gets a paper with her own name on the byline of something she never wrote. A grant reviewer finds her own nonexistent work footnoted back at her. A one-year editor watches phantom references sail through three rounds of peer review and only surface at copyediting. A whistleblowing paper about ethics gets caught faking the citations that were supposed to prove it was rigorous.

The presumption has flipped. A citation used to be innocent until shown otherwise. Now the responsible move, the one editors like Harkins-Brown are already doing by hand, is to assume every reference is fake and ask the author to produce the receipt. That works. It's also slow and manual, and depends on someone remembering to do it before the paper reaches print, which, per the Journal of Academic Ethics, is not guaranteed.

You can't prompt your way out of this. Asking ChatGPT to double-check its own work fails too, the model has no way to know what it invented versus what it recalled correctly, and asking it to grade its own homework just produces a second confident answer. The fix is checking every citation against the actual, live scholarly record before the paper goes anywhere near a reviewer: does this paper actually exist. Boring, mechanical work. It's also the only step in this entire story that would have stopped every single one of these papers before an independent researcher in Finland had to do it for free.