Spotting Fake Citations from AI

A language model writes by predicting which words are likely to come next. Citations follow a very predictable pattern: author names, a year, a title, a journal, a volume, page numbers and often a DOI. A model has seen millions of real references, so it can produce one that looks perfectly formed and points to nothing. This is not the model lying on purpose. Unless it is connected to a search tool, it has no built-in step that checks whether the reference exists. Bad AI citations come in four kinds. Some are completely made up. Some mix real researchers with a title they never wrote. Some name a real paper but get details wrong, such as the year, the journal or the page numbers. The hardest kind to spot is a real, correctly cited paper that doesn't say what the chatbot claims it says. The consequences are real. In the 2023 case Mata v. Avianca in federal court in New York, lawyers filed a brief citing court cases that ChatGPT had invented, and the judge sanctioned them. To check sources before you submit, first search the exact title in quotes in Google Scholar or your library's database. Second, if there is a DOI, open it through doi.org and confirm the paper it leads to has the same title and authors. Third, check that the authors, year and journal all match. Fourth, open the source and find the passage that supports your claim. Fifth, if you can't verify a source, remove it. A common mistake is asking the chatbot to confirm its own citation. It can produce a confident 'yes' just as easily as it produced the fake reference. Tools that search the web lower the risk but don't remove it, because they can still misstate what a real source says.