Paste a bibliography from ChatGPT or any LLM and we cross-check every reference against 6 scholarly databases, resolve each DOI independently, and flag citations that don't hold up.
*Free account, no credit card — 50 credits/month (1 credit per verification, 4 per citation search).
Enter text and use \cite(n) to request n citations. Example: "Deep learning has improved image recognition \cite(3). Furthermore, transformers are now widely used in computer vision \cite(2)."
We found 24 citations. 2 flagged as potential hallucinations.
Uploaded files aren’t stored — they’re processed for this scan and discarded once it completes. See our Privacy Policy.
ChatGPT and other language models don't look papers up when they write a reference — they predict text that statistically resembles a citation, based on patterns learned from real bibliographies. That's why hallucinated references often look perfectly formatted: a plausible author list, a journal name that exists, a DOI-shaped string. But the paper title may not match any real publication, the listed authors may never have written together, or the DOI may resolve to a completely different paper (or nothing at all). These aren't rare edge cases — studies have repeatedly found LLM-generated bibliographies containing fabricated references, which is a serious problem when they end up in a thesis, grant application, or legal brief.
We parse every reference you paste in and cross-reference it against OpenAlex, Crossref, Semantic Scholar, PubMed, DBLP, and arXiv at once, looking for independent agreement between sources rather than trusting the first database that answers. We also separately resolve each DOI and confirm the title actually appears on the page it points to, which catches the common case of a citation with a real-looking but mismatched or stale DOI. If a reference doesn't match closely enough on any of these checks, we flag it as unverified or likely-hallucinated and suggest a real paper to replace it, rather than assuming it's genuine just because it's correctly formatted. We also flag retracted papers using retraction data already published by OpenAlex, Crossref, and PubMed — a citation can be a real, correctly-matched paper and still be one you shouldn't be citing. On our own leak-free benchmark (ground truth sourced independently of every database we check against), this catches 99.4% of fabricated citations at 90.1% overall accuracy — see the full benchmark methodology.
LLMs generate references by predicting plausible-sounding text, not by looking papers up in a database. That produces titles, author lists, and DOIs that read as real but often don't correspond to any actual publication.
We query OpenAlex, Crossref, Semantic Scholar, PubMed, DBLP, and arXiv, plus live URL checks for non-academic web sources, and look for independent agreement between them.
We flag it as an unverified or likely-hallucinated citation and suggest a real paper to replace it, instead of assuming it's genuine because it looks correctly formatted. On our own leak-free benchmark, this catches 99.4% of fabricated citations — see the full methodology.
Yes — a real, correctly-matched citation can still be flagged as retracted, separately from whether it verifies. We check retraction metadata already published by OpenAlex, Crossref, and PubMed, so a paper doesn't have to fail our match to be worth a second look.
Guides on catching hallucinated citations, comparisons of the tools available, and case studies from peer review and grading.
AccuraCite Now Has an MCP Server
AccuraCite now ships an MCP server for Claude, Cursor, and other MCP clients — the consensus verification engine, not a single tool call, exposed over MCP.
We Benchmarked AccuraCite's Citation Finder Against Raw AI — Here's What We Found
We tested AccuraCite's Find Citations feature against a naive AI answer and a naive title match, on claims about obscure, recent research. The grounded pipeline won by a wide margin.
What Is a Retracted Citation, and Why 'Verified' Doesn't Mean Safe to Cite
A citation can be 100% real, correctly matched, and still be a paper you shouldn't cite. Here's what a retraction is, how AccuraCite detects it across three databases, and two real cases that prove it works.
What Is DOI Verification, and Why a Real DOI Isn't Enough
A DOI that resolves isn't proof a citation is real. Here's what DOI verification actually checks, the ways a DOI can be wrong, and how to check one properly.
AccuraCite Catches 99.4% of Fake Citations — Here's the Leak-Free Benchmark
We tested AccuraCite against 918 real citations, including citations built specifically to be hard to catch. It caught 99.4% of the fakes.
How to Check Citations as a Peer Reviewer (Before You Sign Off)
Reviewers rarely have time to check 60 references by hand. Here's how to bulk-verify a manuscript's citations before you submit your review.
The Best AI Citation Checkers in 2026: What We Could Actually Verify
We compared AccuraCite, Citely, GPTZero, and CiteTrue as citation checkers, verifying vendor claims directly instead of repeating unconfirmed numbers.
How to Find Sources for a Claim When AI Can't Back It Up
Asking ChatGPT for a source often returns a hallucinated citation or a real paper that doesn't back your claim. Here's how to find sources that actually do.
Why AI Citation Checkers Get Fooled by Fake DOIs (And How We Fixed Ours)
A citation-verification tool told us a famous 2017 paper was published in 2025 by a fake publisher. Here's the real bug, and the fix AccuraCite shipped.
Seven Fake DOIs, One Famous Paper: A Warning for Journal Editors
A real duplicate DOI impersonates a famous 2017 paper seven times over in Crossref. What it means for verifying manuscript references before publication.
How to Check If ChatGPT's Citations Are Real Before You Submit
Did ChatGPT give you real sources? Here's how to check every citation in your own paper before you submit it, in one pass instead of Googling each by hand.
Why 'Just Use an LLM with MCP Tools' Doesn't Solve Citation Hallucination
MCP lets an LLM call academic APIs directly. That sounds like it solves citation hallucination. It doesn't, by itself — here's the actual failure mode, with a real example.
How to Check Your Students' Citations for AI Hallucinations
Turnitin catches copied text, not fake sources. Here's how to bulk-check student papers for AI-hallucinated citations in one pass, without manually looking up every reference.
Verify your bibliography — free, in seconds.
Create a free account to get 50 credits/month (1 credit per verification, 4 per citation search).
Log In / Sign Up