No. Language models routinely fabricate citations that look completely plausible — real author, real journal, realistic title, a volume and page range — for an article that was never written.
Submitting one claims you read something you didn't, which most universities treat more seriously than the AI use itself.
Why fabricated references look so convincing
A model isn't looking anything up. It's generating the text most likely to follow your request, and it has seen enormous numbers of real citations — so it knows exactly what one looks like. The output is a perfectly-shaped reference: plausible author for that field, real journal, credible year, sensible page range. Everything correct except that the article doesn't exist.
This is why the usual sanity checks fail. It doesn't look wrong. It looks like every other entry in your list.
A real paper attributed to the wrong author. A real author credited with a paper they didn't write. A real article with the year or volume altered. These survive a casual search — you find something — and only fall apart when a marker checks the actual record.
Why it's treated seriously
A fabricated citation isn't a formatting error. It's a claim about your research: that you located, read and used a source. If the source doesn't exist, the claim is false regardless of whether you meant it — and most misconduct regulations have wording for that which predates AI entirely.
It's also the easiest thing in the world for a marker to check. One search. Academics in a field recognise their own literature, and a confidently cited paper that nobody has heard of stands out immediately. See is using AI to write an essay plagiarism? for how the categories fit together.
How to verify a reference properly
- Find the actual record. Your library search, Google Scholar, or the publisher's own site. Not a summary of it, not a citation of it elsewhere — the record itself.
- Check the DOI resolves. Paste it after
https://doi.org/. A fabricated DOI goes nowhere, and this is the fastest single test. - Match every field. Author spelling, year, volume, issue, pages. Half-real references fail here.
- Read enough to cite it honestly. If you only read the abstract, don't cite it as though you read the argument.
Reference tools that retrieve real records — your library database, a reference manager — are a different category from a chatbot and are fine to rely on, because they're looking things up rather than generating text.
Using AI around references safely
Asking a model to format a reference you already have is low-risk: the source exists and you're only rearranging fields, though you should still check it against your department's guide. Asking it to find sources is the dangerous direction. And if you cite an AI conversation itself, there's a right way to do it — see how to reference ChatGPT in your essay.
SafeGrade checks Harvard and APA formatting, flags uncited entries and inconsistencies, and shows you exactly which line to fix. Never stored, never used for training.
Run one free scan →No card. No catch. First scan free.