Citation verification glossary
Plain-English definitions of the identifiers, databases, and failure modes behind checking whether a citation is real.
- DOI (Digital Object Identifier)
A DOI is a permanent identifier assigned to a scholarly document, formatted like 10.1038/nature14539 and resolvable at doi.org.
Because DOIs are registered, a citation with a DOI can be checked directly against the registry — but a real DOI can still be attached to the wrong paper, which is a common AI-fabrication pattern.
- PMID (PubMed ID)
A PMID is the unique number PubMed assigns to a biomedical article, used to look it up in the National Library of Medicine's database.
Vancouver-style biomedical citations often carry a PMID. A fabricated PMID typically resolves to an unrelated paper.
- arXiv ID
An arXiv ID identifies a preprint on arXiv.org, written like arXiv:1706.03762, where physics, math, and computer-science papers are posted before peer review.
An arXiv ID attached to the wrong title is a classic failure mode — the identifier looks authoritative even when it names a different paper.
- ISBN
An ISBN is the 10- or 13-digit identifier for a published book edition, used to verify a book's existence and metadata.
Books without an indexed ISBN match are why some citations return “can't check” rather than a confirmation — absence of an index entry is not proof a book is fake.
- Citation hallucination
A citation hallucination is a reference invented by an AI language model that looks correctly formatted but does not correspond to any real source.
Language models generate text that looks right; a citation is a highly regular pattern, so a model can produce a perfect-looking one with no underlying paper. Verifying against authoritative databases is the only reliable way to catch these.
- Retraction
A retraction is the formal withdrawal of a published paper by its journal, usually for fraud, fatal errors, or misconduct, signalling the work should no longer be relied upon.
Retracted papers keep getting cited for years afterward. Retraction status comes from sources like Retraction Watch, Crossref and OpenAlex and should be surfaced before a paper is cited as evidence.
- Preprint
A preprint is a version of a paper shared publicly (e.g. on arXiv or bioRxiv) before peer review and formal publication.
A preprint and its later published version are the same work in two forms; a good check links them rather than treating a year difference as an error.
- Predatory / low-trust journal
A predatory journal publishes for a fee without genuine peer review or editorial standards, lending fabricated-looking work a veneer of legitimacy.
A registered identifier alone does not prove a source is trustworthy — existence and metadata can check out even when the venue's quality does not.
- Crossref
Crossref is the largest scholarly metadata registry, indexing 160M+ works and the DOIs assigned to them.
It is the backbone source for citation verification, queried by bibliographic string or by DOI.
- OpenAlex
OpenAlex is a free, open index of scholarly works, authors, and venues, and a superset of Crossref that also carries retraction and open-access flags.
- Bibliography vs. reference list
A reference list contains only the sources actually cited in a work; a bibliography may also include background reading that was not directly cited.
- et al.
“Et al.” is Latin for “and others,” used to shorten a citation with many authors after the first one or few are named.
Dropping the real first authors of a paper while keeping later ones is a subtle citation error that field-level checking can catch.