Fake Citation Detector — Flag Hallucinated Papers in Seconds
A fake citation points to a paper that doesn't exist — or to a real paper wearing the wrong metadata. EvalCite detects them by checking every reference against 8 scholarly databases and diffing each field against the authoritative record. Free, no signup, results in seconds per reference.
A taxonomy of fake citations
Chimera citations
Named for the mythological hybrid: the head of one paper, the body of another. A real, prominent author's name attached to a title they never wrote, published "in" a journal they genuinely publish in. Chimeras survive casual inspection because every individual element looks right — only a database lookup reveals the combination doesn't exist.
Fabricated DOIs
A DOI with the right syntax (10.xxxx/…) that resolves to nothing — or worse, resolves to a completely different paper. EvalCite resolves the DOI directly and compares the record it returns against the rest of the citation.
Metadata drift
The paper is real, but the year is off by two, the volume is wrong, the pages belong to a different article, or the author order is shuffled. Drifted citations often come from AI paraphrasing of a real source — or from sloppy copying. They surface as mismatch with the exact disagreeing fields highlighted.
Ghost papers
Entirely fabricated: a plausible title, real-sounding authors, a real venue, specific page numbers — and no record anywhere. Ghosts are the most common product of LLM "research assistance" and the easiest for EvalCite to flag: nothing in 8 databases comes close.
How detection works
- The citation is parsed into title, authors, year, venue, and identifiers.
- Exact identifiers (DOI, arXiv ID) are resolved first — a resolving DOI with mismatched metadata is flagged immediately.
- Otherwise, all databases are searched concurrently and the best candidate is matched on title and authors, with a year check.
- A confident match → verified; a partial match → mismatch with a field-level diff; nothing close → not found.
- If the citation carries a URL no database knows, the live page is fetched and its title matched — this rescues real blog posts and model cards without letting fabricated titles at real URLs through.
Frequently asked questions
Can a real paper be flagged as fake?
Occasionally: very recent papers not yet indexed, and non-scholarly sources without URLs, can read not found. That's why the verdict is "not found" rather than "fake" — treat it as a prompt to check manually, not a conviction.
Does it detect fake DOIs specifically?
Yes. DOIs are resolved against CrossRef; an unresolvable DOI or one whose record doesn't match the cited title/authors is flagged.
What about predatory-journal papers?
EvalCite verifies existence and metadata, not journal quality. A real paper in a predatory journal verifies — quality judgment stays with you.
How is this different from a plagiarism checker?
Plagiarism checkers compare prose against published text. EvalCite checks whether the references themselves point to real publications — a different, complementary check.