Do your references actually exist?

What the verdicts mean

Verified

The identifier (DOI, arXiv ID or PMID) resolves, and the registered title, author and year match this citation.

Match found

No identifier was given, but a registry record closely matches the cited title, author and year.

Exists

A bare identifier with no citation text around it — it resolves, but there is nothing to compare it against.

Check me

A record was found but only partly matches (e.g. right title, wrong year). Compare by hand before trusting either way.

Unconfirmed

The DOI is registered (per doi.org) with an agency whose metadata this tool cannot read — existence confirmed, contents not.

Mismatch

The identifier is real but registered to a different work than the one cited — the classic AI-fabrication signature.

Not found

The identifier is definitively absent from the global registries (confirmed 404, not a network hiccup).

Not indexed

No identifier, and no close match in the indexes searched. Could be fabricated, or simply not indexed (books, theses, non-English, very new). The tool can't tell which — a lead to check, never a conviction. Only the two red verdicts are things it can prove.

Retracted

The work exists but has been retracted (per OpenAlex). Citing it as valid evidence is misleading.

Can't check

Web pages, news articles and other non-scholarly sources can't be fetched from your browser. Open the link and check it yourself.

Retry

A registry could not be reached. This is explicitly not a verdict about the reference.

What this tool does not tell you

• “Verified” means the reference exists — not that it supports the claim it is cited for. A real paper can still be cited misleadingly. Read your sources.
• No legal case-law. Court citations (cases, statutes) live in different databases. For US case law try CourtListener.
• Coverage has edges. Books, theses, reports, non-English and very recent works are under-indexed. “No record” for these is weak evidence — verify by hand.
• Retraction data is incomplete. A missing retraction flag is not proof a work stands; check Retraction Watch for anything critical.
• Match scoring is heuristic. The similarity between your citation and the registry record is computed from titles, authors and years — it can misjudge unusual formats, and references whose first author has a very short surname may land on “Check me” rather than green. The evidence for every verdict is shown so you can overrule it.
• Paste plain references. One per line, per blank-line block, or numbered ([1], 1.). BibTeX and RIS files aren't parsed yet — paste the human-readable reference list, not the .bib.

Privacy, by construction

There is no server behind this page — it is a static file. Your text is never uploaded to us because there is no “us” to upload to. When you press Check, your browser queries the public scholarly registries directly (Crossref, DataCite, doi.org, OpenAlex, PubMed, Open Library), exactly as if you had searched them yourself. Those registries see the individual reference queries — from your IP address, not through any intermediary. The Crossref queries also carry a fixed contact address (ghostcite@dsl4.com, the same for every visitor, per Crossref's politeness convention) — that identifies the tool, never you. No account, no cookies, no analytics, no logs of your text. The page's Content-Security-Policy makes this enforceable: the browser will not load a script, or open an app connection, to anywhere but those six registries. (The host, Cloudflare, tries to inject its own cookieless analytics beacon, as it does on every site it serves; this page's policy blocks it before it runs, so no analytics loads — you can confirm that in your browser's network tab.) The entire checker is one readable JavaScript file — audit it.