AI Citation Checker — Detect Fake References from ChatGPT & Claude

Large language models fabricate academic references: real-sounding titles, plausible authors, journals that exist — for papers that don't. EvalCite checks every reference against 8 scholarly databases (CrossRef, OpenAlex, Semantic Scholar, DBLP, arXiv, OpenLibrary, Google Books, plus IEEE Xplore and Scopus with keys) and flags the ones no database can confirm. Free, no signup, no character limits.

Why AI chatbots invent citations

LLMs generate text by predicting plausible continuations, not by looking papers up. Asked for references, a model reproduces the shape of a citation — a title in the field's style, well-known author names, a real journal, a year — without any underlying record. Audits of AI-assisted manuscripts consistently find fabricated or corrupted references, and reviewers increasingly check bibliographies before reading a word of the argument.

How EvalCite catches them

1. Parse

Your list is split into individual references and each is parsed into structured fields — title, authors, year, venue, DOI, arXiv ID. Numbered lists, author-year (APA/Chicago) lists, and BibTeX files are all handled.

2. Verify

Each reference is looked up concurrently in all 8 databases and the best-matching record wins.

3. Verdict

Verified means a real record matches. Mismatch means a similar paper exists but details differ — the classic signature of an AI citing a real paper with invented metadata. Not found means nothing close exists in any database: for AI-generated text, that almost always means fabrication.

The patterns we see in AI-hallucinated references

• Chimera citations — a real author's name attached to a title that author never wrote, in a journal they publish in.
• Fabricated DOIs — DOIs that follow the publisher's syntax but resolve to nothing (or to a different paper).
• Metadata drift — a real paper with the wrong year, volume, pages, or author order.
• Ghost papers — entirely invented titles in real venues, often with impressively specific page numbers.

Frequently asked questions