The traditional time to check the references
Check your references before the portal does.
This is naturally when we discover that one DOI opens the wrong paper, another appears twice, and the preprint may have grown into a publication.
Paste the list. The tool checks whether each identifier resolves, compares the title, authors, journal, and year with what you wrote, and lists back the references it could not check. It cannot tell whether the source supports your claim. Nothing is stored; queries go to Crossref, DataCite and OpenLibrary for the record, and to Unpaywall for open-access status.
Queries go to Crossref for the record, to DataCite when Crossref has no entry (preprints, datasets, arXiv), to OpenLibrary for an ISBN, and — while the box above is ticked — to Unpaywall for access status. Nothing else leaves the browser, nothing is stored, no account, no telemetry.
What this checks
- Does the identifier resolve? Each DOI is looked up in Crossref, and in DataCite when Crossref has no entry — preprints, datasets and software register there, and an arXiv ID resolves through its own DataCite DOI. An ISBN is looked up in OpenLibrary. Found, or not found.
- Does the record match your citation? Four fields are compared with the citation text around the DOI: the title, the author list (in order, with "et al." expected), the journal, and the year. Journal names are matched abbreviated or in full, so "N Engl J Med" and "The New England Journal of Medicine" both count. Each field that doesn't line up is flagged on its own, so you can see which one is off.
- Does your citation name someone who isn't on the paper? The surnames in your citation are read back and checked against the record's whole author list. A name the record doesn't carry anywhere is flagged wherever it sits. This is where a re-typed or machine-repaired list usually goes wrong: the first author survives, and a name in the middle quietly does not.
- What it couldn't check. Entries that read like references but carry no DOI, arXiv ID or ISBN are listed back to you, unchecked, and counted separately from the ones that were. The tool can't look a reference up by title, so those stay yours.
- Take the result with you. Copy the run as Markdown, or download it as a .md file: every reference, what the record said, which fields didn't line up, and what wasn't checked. It pastes into a manuscript file or an email to a co-author, and it carries its own limits, because a file travels away from the page that explains it.
- Retractions and corrections: flagged from Crossref update metadata, naming Crossref as the source.
- Preprint DOIs: labelled preprint, so you can check for a published version. A preprint is not a problem.
- Malformed DOIs and duplicates: a DOI that isn't well-formed, or the same DOI cited twice.
- Open access or paywall? Each resolved DOI is checked against Unpaywall for a free full-text copy. When one exists, the tool shows the best single-hop link — a PubMed Central link where Unpaywall lists one — so you can prefer it over a DOI that may stall behind a redirect. Access status sits beside the verdict and never changes it: a paywalled paper is not a failure.
Every reference gets one of three verdicts: checks out (resolves and matches), does not check out (resolves but the record disagrees, or is flagged), or not resolvable (no registry returned a record, so there is nothing to compare). A missing record is never counted as a failure. The verdict answers one question: does this DOI point at the paper you cited? A journal, author, or year that doesn't line up is flagged beside the verdict rather than inside it, because a typo in the reference is a different problem from a wrong DOI.
Checking the wording of a paragraph instead of the reference list? Try the Citation Integrity Quick Check.