How to Spot Fake AI (ChatGPT) Citations

An AI-hallucinated citation is a reference produced by a language model that does not correspond to any real publication. The author names may be real, the journal may exist, and the title may sound exactly like a paper that should have been written — but the specific work was never published.

This happens because language models generate text by predicting what is statistically likely to come next, not by looking anything up. When you ask for sources, the model produces text shaped like a citation. Unless it was connected to a real database at the moment it answered, it has no way to know whether that citation exists.

Hallucinated references are dangerous precisely because they are well-formed. A model trained on millions of real citations reproduces the genre almost perfectly:

• Real, recognisable author names — often leading researchers in the field.
• Real journals with plausible volume and issue numbers.
• A title that fits your topic so well it feels like it must exist.
• A complete, correctly formatted DOI — that resolves to nothing.

• The DOI returns "DOI Not Found" at doi.org.
• The exact title, in quotes, returns no results in Google Scholar, CrossRef, or OpenAlex.
• The authors are real but never co-published this title, or never worked on this topic.
• The year is impossible — before the journal launched, or after a researcher stopped publishing.
• The paper is cited nowhere else and has no abstract, no PDF, and no publisher page.

Eyeballing each reference works but does not scale, and the convincing ones are exactly the ones you will miss. The reliable approach is to verify every reference against the real literature — match each title, author set, year, and DOI against databases of hundreds of millions of works, and flag anything with no match.

That is what an AI reference verifier does: paste the citations a chatbot gave you, and it checks each one against real academic sources, marking the ones it cannot confirm so you can replace them before submitting.

• Never paste a chatbot's citations into a paper without verifying them first.
• When a reference fails, regenerate it from a real DOI, URL, title, or ISBN — do not edit the fabricated text.
• Keep the verified source, not the AI's wording: open the real paper and cite what it actually says.