The process
How citation verification works
Verification checks the record. Human review checks the meaning. Here is exactly what happens between pasting a citation and seeing a result.
1. Your citation is parsed into structured fields
The raw text is normalized — whitespace, punctuation, Unicode — and split into the fields a record can be matched on: title, authors, year, journal or venue, volume, issue, and pages. At the same time, the text is scanned for identifiers: DOIs, PubMed IDs, and arXiv IDs. Parsing is deterministic; no language model rewrites or "fixes" your citation.
2. Identifiers are resolved first
An identifier is the strongest evidence a citation can carry, so it is checked before any fuzzy matching. A DOI is looked up at Crossref (with OpenAlex as a fallback), a PMID at PubMed, an arXiv ID at arXiv. If the identifier resolves, the returned record becomes the comparison target — including when the record it resolves to is not the paper the citation describes. A real DOI attached to the wrong title is one of the most common problems we see, and it is treated as a partial match, not a pass.
3. Without an identifier, records are searched
The parsed fields are sent as search queries to scholarly metadata providers — currently Crossref and OpenAlex for general search. The top candidates come back and each is compared with your citation. When several records are plausible, alternatives are shown rather than silently picking one.
4. Fields are compared, one by one
Titles are compared after normalization (casing, punctuation, accents) so that trivial differences don't count against a match — but meaningful differences are preserved. Author family names are checked for overlap. Years may differ by a small tolerance, because online-first publication and indexing genuinely shift years. Venues, volumes, issues, and pages are compared where present.
5. The result is classified — and explained
Documented thresholds convert the field comparisons into one of five states: Verified, Partial match, Needs review, Not found, or Service unavailable. The thresholds live in configuration and are described on the methodology page. Every result shows the evidence behind it: the matched record, the field comparison, the sources checked, and any provider problems.
What the states mean
What each result means
-
Verified
The citation appears to correspond to a real record, and its important details substantially agree. It does not mean the source is high quality, unretracted, or that it supports your claim.
-
Partial match
A likely related record exists, but one or more important fields differ — a wrong year, different authors, or a DOI pointing at a different paper.
-
Needs review
The evidence is incomplete, ambiguous, or conflicting. A human needs to look at the source.
-
Not found
The enabled sources returned no adequate record. Not found does not automatically mean fake — coverage, spelling, indexing delays, and non-scholarly sources all affect results.
-
Service unavailable
A data provider failed or timed out, so the check could not be completed. This says nothing about the citation itself.
Two warnings worth repeating
Not found does not automatically mean fake. Coverage, spelling, indexing delays, non-scholarly sources, and incomplete metadata can all cause a real work to go unfound.
Verified does not mean good. A verified record can still be weak, irrelevant, corrected, retracted, or misused. Always review the source itself.
Check the citation before you trust it.
Free citation checks with visible source evidence. No account needed for basic checks.
Check citations — free