How to Spot a Fake AI Citation (Before It Costs You Marks)

Ask ChatGPT for sources and it will hand you a beautifully formatted reference list — correct author names, a real-sounding journal, a tidy DOI. The problem: a striking share of those citations point to papers that were never written. The model isn't looking anything up; it's predicting text that looks like a citation. Studies have found that more than half of AI-generated references are fabricated or wrong — and turning one in can cost you marks or trigger an academic-integrity case.

Why AI makes up citations

A large language model generates the most plausible next words, not verified facts. Unless it's connected to a real database, a “citation” is just plausible-looking text: a realistic author, a real journal name, and an invented title glued together. That's why fake citations are so convincing — every part is realistic, even though the whole doesn't exist.

Five signs a citation might be fake

1. The DOI doesn't resolve. Paste it after https://doi.org/ — a real DOI lands on the paper; a fake one 404s.
2. You can't find it. Search the exact title in quotes on Google Scholar or the journal's site. A real paper shows up; a fabricated one returns nothing.
3. Real authors, wrong pairing. The names are real researchers who never co-wrote this — or never worked on this topic.
4. The details don't line up. The volume/issue/year/page numbers are inconsistent (e.g. a volume that journal hadn't reached in that year).
5. It's suspiciously perfect. Exactly on-topic, round page numbers, no messy metadata — real bibliographic data is rarely this clean.

The 60-second manual check

For each suspicious reference: (1) copy the DOI and open doi.org/<doi>; (2) search the exact title in quotes on Google Scholar; (3) open the journal's own site and confirm the volume and year. If any of those fail, treat the source as unverified and don't cite it.

The faster way: check them all at once

Doing that by hand for a whole bibliography is slow. Paste your reference list into Corroba Verify and it checks every entry against real scholarly databases (Crossref, OpenAlex, PubMed) — flagging broken DOIs, retracted papers, and any source it can't confirm exists, and showing you the real record when there is one. It even grades each source's quality (peer-reviewed vs. preprint vs. web). It never invents a source to fill a gap — if it can't be verified, it says so.

Wrote your draft with an AI assistant? The AI citation checker is built for exactly that. And if you need to cite the AI itself, here's how to cite ChatGPT correctly.

FAQ

Does ChatGPT really make up citations? Yes — routinely. Even when the authors and journal are real, the specific paper often isn't. Always verify.

Will my professor notice? Increasingly, yes — a citation that doesn't resolve or can't be found is a red flag, and many instructors now spot-check references.

How do I fix a fake citation? Find a real source that actually supports your point and cite that instead — never keep a reference you couldn't verify.