Citation Verifier

Each reference is checked with a bibliographic search against Crossref and Semantic Scholar, a cleaned term search against PubMed, and a full-text search against arXiv (volume/page/date noise is stripped first for PubMed and arXiv, since raw citation punctuation can trip up their query parsers). Each source returns its top few candidates, not just its #1 ranked guess, and the tool scores every candidate's title against your pasted text — this catches cases where the right paper wasn't ranked first. If a DOI is present in the pasted text, it's confirmed directly against Crossref's registry first and treated as authoritative; DOIs with the 10.1101/ prefix are also confirmed directly against the bioRxiv and medRxiv preprint APIs.

• Verified — DOI confirmed (including via bioRxiv/medRxiv), or ≥75% word overlap with a real indexed record.
• Possible match — 40–75% overlap. Worth a manual glance; could be a paraphrased title, a preprint vs. published version, or a partial citation.
• No match / check failed — either no record cleared 40% overlap in any of the four searchable indexes, or at least two of them couldn't be reached (network issue or rate limit) — both are folded into one neutral count so an unreachable API is never confused with a real non-match. Each card says which case it is, and has a recheck button.

Google Scholar isn't included — it has no public API, and scraping it violates its terms of service and gets blocked by CAPTCHA in practice. bioRxiv/medRxiv's own search isn't included either, for a related reason: their public API only supports direct DOI lookups, not free-text search, so bioRxiv/medRxiv coverage here is DOI-only. Crossref, PubMed, Semantic Scholar, and arXiv together cover the large majority of what Scholar indexes for genuinely published or preprinted work, and return structured metadata rather than scraped HTML, which is what makes this kind of automated check possible at all.

A "no match" is a signal to check by hand, never proof of fabrication. Very new papers, non-English titles, book chapters, conference proceedings, and paywalled or niche sources can all be under-indexed or absent from these databases even when they're completely real. Semantic Scholar's free public API in particular has a low, shared rate limit and can fail intermittently under load — that's what the recheck button is for. Treat this tool as a fast triage pass that narrows down what to check by hand, not a final verdict.

All lookups run directly from your browser to Crossref, PubMed, Semantic Scholar, arXiv, and the bioRxiv/medRxiv APIs — nothing is sent anywhere else or stored.

Filtering and export: click any count in the summary row to show only references in that category; click it again (or "show all") to clear the filter. Verified references can be reformatted into APA, MLA, Chicago, Harvard, or Vancouver style and exported as BibTeX or RIS. RIS is a shared, well-documented format that EndNote, Reference Manager, RefWorks, Zotero, and Mendeley all import directly — rather than three separate near-identical proprietary exports, one RIS file works across all of them. Formatting relies on whatever metadata the matched record provides (author list, year, journal, volume/issue/pages, DOI); fields the source didn't return are simply omitted, so it's worth a final check against your target style guide before submission.