How do I spot a hallucinated reference?

Search for it in an indexed database or by its DOI. Warning signs include a DOI that does not resolve, a title that does not appear anywhere, a real author paired with a paper they never wrote, a journal that does not carry that article, and details that shift when you ask the model again. If you cannot independently find the source, treat it as fabricated.

Is using a fabricated AI citation misconduct?

Citing a source you have not verified, that turns out not to exist, can amount to fabrication or at least serious negligence, regardless of whether an AI produced it. You remain responsible for every citation you submit. The protection is simple: never cite anything you have not personally located and read.

Home › Guides › AI in Research › AI hallucinations

AI hallucinations & fabricated citations

Q: What is an AI hallucination?

An AI hallucination is confident output that is false or fabricated, such as an invented citation, a misattributed quote, or a made-up statistic. The model is not lying in any intentional sense; it generates the most plausible-sounding text, which is not the same as the true text. In research the most damaging form is the fabricated reference.

Q: Why does AI make up citations?

A language model predicts likely sequences of words rather than retrieving records from a database. A citation has a predictable shape (authors, year, title, journal, DOI), so the model can generate one that looks completely real without any underlying source existing. That is why fabricated references often combine plausible authors and journals that never published the cited paper.

By Dr. Rafiq Muhammad, MD, PhD · Updated June 2026

The most dangerous thing AI does in research isn’t getting things wrong — it’s getting things wrong confidently, in the exact format of a real citation. A “hallucination” is plausible output with no basis in fact: an invented reference, a misattributed quote, a made-up statistic. Knowing why it happens is what lets you catch every instance before it reaches a reviewer.

Real cases exist of researchers and lawyers submitting work with citations that simply don’t exist. The fix is not better AI — it’s a verification habit you never skip.

Why models invent citations

A language model predicts likely text; it doesn’t look anything up. A citation has a highly predictable shape — authors, year, title, journal, DOI — so the model can assemble one that looks perfect without any real paper behind it. That’s why fabricated references so often pair a plausible-sounding author with a journal that never published the work, or wrap a real title around a wrong author and year. It’s not lying; it’s pattern-completion mistaken for retrieval.

How to spot a hallucinated reference

The DOI doesn’t resolve — paste it into doi.org and you get nothing.
The title returns no hits in Google Scholar or the database it claims to be in.
A real author, a phantom paper — the author exists but never wrote that.
Wrong journal — the journal is real but doesn’t carry the article.
Details drift — ask again and the year, pages, or authors change.

If you can’t independently locate the source, treat it as fabricated — full stop.

A verification routine that never fails you

Never cite what you haven’t read. This single rule eliminates the entire risk.
Check every reference against an indexed database or by DOI before it enters your reference list.
Trace every fact and number back to a primary source — not the AI’s paraphrase.
Be just as wary of summaries. A hallucination can be an inverted finding, not just a fake citation.

To build a reference list from real indexed work rather than invented citations, the free Research Assistant points you to the actual literature for your topic — sources you can open and verify.

Get the free AI-in-Research Ethics & Disclosure Pack

An AI-output verification checklist, disclosure-statement templates, and an acceptable-use guide from The AI-Powered Scholar. We’ll email you the download link.

Get The AI-Powered Scholar

Frequently asked questions

What is an AI hallucination?

Confident output that’s false or fabricated — an invented citation, misattributed quote, or made-up statistic.

Why does AI make up citations?

It predicts plausible text rather than retrieving records, and a citation’s predictable shape is easy to fake convincingly.

How do I spot one?

Search it in an indexed database or by DOI. Non-resolving DOIs, no title hits, mismatched author/journal, or shifting details all signal fabrication.

Is citing a fake AI reference misconduct?

Citing an unverified, non-existent source can be fabrication or serious negligence — you’re responsible regardless of the AI. Never cite what you haven’t read.

AI ethics & disclosure → Open the Research Assistant →