Reference Checker

Advertisement

Reference Checker

Paste a reference list (a bibliography, a BibTeX or RIS file, or a list an AI assistant produced) and every entry is checked against the scholarly registries. Only the references are sent; nothing else leaves the browser.

    Cleaned reference list?Each entry uses your choice per row: the registry record or the reference as pasted. Nothing is dropped.

    Was this tool helpful?
    Want to tell us more?
    0/500
    Want us to follow up?
    Thanks for your feedback!
    Advertisement

    About This Tool

    The checker takes a whole reference list at once and verifies every entry against the registries publishers themselves maintain: Crossref for journal articles and books, Europe PMC for PubMed records, DataCite for datasets and preprints, and the book catalogs for ISBNs. A reference that carries an identifier (a DOI, PMID, ISBN, or arXiv ID) is resolved directly. A plain-text reference is searched by its title, authors, and year, and the best match is scored before it is trusted. Each entry answers one question: does this reference point to a real, findable record, and do its details agree with that record?

    How to Use

    1. Paste the reference list — numbered, bulleted, BibTeX, or RIS.
    2. Select Check references and follow each entry as its state arrives.
    3. Review the differences, pick record or pasted wording per row, and copy or export the cleaned list.

    Methodology

    Identifier lookups come first: a DOI resolves through doi.org content negotiation, a PMID through Europe PMC, an ISBN through Google Books with an Open Library fallback, and an arXiv ID through its DataCite record. For an identifier, the registry record is treated as authoritative. Text-only references go through Crossref bibliographic search. The best candidate is compared with the pasted text on title, authors, and year; only a strong agreement is accepted, and partial agreement is marked for review with each differing field listed. Absence from the indexes is reported as an open question, not as proof a source is fabricated: many books, chapters, and older or regional venues have no DOI. Such entries are marked "could not verify" and kept exactly as pasted.

    Sources: Crossref REST API · Europe PMC REST API · DataCite REST API

    Understanding Your Results

    "Verified identifier" means the reference carries a DOI, PMID, ISBN, or arXiv ID that resolved to a registry record; listed differences compare your text against that record. "High-confidence match" means a text search found a record agreeing on title, authors, and year. "Possible match" means one record came close but at least one field disagrees; read the listed differences and decide. "Could not verify" means no convincing record was found. For a list an AI assistant produced, that is a strong warning sign; on its own, it is not proof of nonexistence. Duplicate flags mark rows that resolve to the same work, which usually means one source was cited twice in different wordings.

    Practical Examples

    A pasted list of twelve references from a draft returns nine verified identifiers, one high-confidence match, one review case where the pasted year says 2004 but the registry record says 2003, and one entry no index can confirm. A list produced by an AI assistant is a typical second case: within seconds, the checker confirms the real entries and leaves the invented ones at "could not verify", before any of them reach a submitted manuscript.

    Tips for reliable checking

    • Keep identifiers in the list: a reference with its DOI verifies directly and precisely. • Check AI-produced lists before any entry reaches your manuscript; invented references usually surface as "could not verify". • Read every review case: a one-year difference is often a citation error worth fixing from the registry record. • "Could not verify" for a book or an older article can simply mean the venue has no DOI; confirm such entries by hand. • Use the duplicate flags to merge double citations of the same work before submission.

    Frequently Asked Questions
    Which registries does the checker use?
    Crossref for journal articles, books, and chapters; Europe PMC for PubMed records; DataCite for datasets, preprints, and arXiv papers; and Google Books with Open Library as the fallback for ISBNs. These are the registries where publishers deposit their own metadata, so a match is evidence from the source of record.
    Can it detect references invented by AI assistants?
    That is its main use. An invented reference has no registry record, so it surfaces as "could not verify"; a real one either resolves through its identifier or matches a record on title, authors, and year. Rather than a verdict of "fake", the tool reports evidence. A missing record can also mean a book or venue without a DOI, so unverified entries deserve a manual look.
    What do the four verification states mean?
    "Verified identifier": the entry's DOI, PMID, ISBN, or arXiv ID resolved to a registry record. "High-confidence match": a text search found a record agreeing on title, authors, and year. "Possible match": one candidate is close but at least one field differs, and each difference is listed. "Could not verify": no convincing record was found; the entry is kept exactly as pasted.
    Which input formats are accepted?
    Numbered and bulleted lists, one reference per line, blank-line-separated lists, BibTeX entries, and RIS records — mixed freely in one paste. Wrapped lines from a PDF copy are joined back together, and list markers such as "1." or "[12]" are removed before checking.
    Is my reference list private?
    Each lookup sends only the single reference being checked to the public registry interfaces: no document, and no full list in one request. Nothing is stored on a server. Results live in the page; saving to My References stores them in the browser only.
    Why is a real book marked "could not verify"?
    Registry coverage of DOIs, PubMed, DataCite, and the book catalogs is extensive, not complete. Books without an ISBN in the catalogs, chapters cited without a DOI, theses, reports, and older or regional venues can be absent. For that reason the state says "could not verify" rather than claiming the source does not exist. Adding an ISBN or DOI to the entry usually resolves it.
    What does the cleaned list contain?
    Every reference, formatted and ordered by the selected style: verified and high-confidence rows use the registry record by default, review and unverified rows keep your pasted wording, and each row's choice can be flipped. Copy the list with formatting, or export it as BibTeX, RIS, or CSL-JSON; the report file records every state and difference.