Before you upload a manuscript, every reference in the bibliography needs to survive three questions: does the DOI resolve, is the paper retracted, and does the metadata you cited actually match the registry? Doing that by hand across Crossref, PubMed, and Retraction Watch takes ninety minutes on a sixty-reference paper — and still misses retractions issued after you last downloaded the PDF.
Cite Trail runs the same checklist in about a minute. Every finding on this page carries a live link to the Crossref, OpenAlex, or Wikidata record it came from. If a reference cannot be verified, we say “no match” instead of guessing. The audit is designed to be defensible in front of a reviewer, not just quick.Sources for every claim on this page live in the audit output — not the marketing copy.
Paste your Methods section. The lint flags factual sentences that lack an inline citation. Every flag is a suggestion for human review — not an assertion of fact.
1. Parse.Your BibTeX, RIS, DOI list, or References-section text is normalized into a list of citation records. When a record has no DOI, we mark it “unverifiable” up front — older papers, book chapters, and preprints often lack a modern registry entry and shouldn’t be silently passed as verified.
2. Crossref. Each DOI is queried against Crossref’s REST API. The update-toarray indicates whether Crossref has recorded a retraction; since Crossref’s 2023 acquisition of the Retraction Watch database this is the closest thing to a canonical retraction feed.
3. OpenAlex fallback. If Crossref returns nothing (common for older or non-DOI-registered work), OpenAlex is queried. OpenAlex’s is_retracted flag adds a second signal.
4. Wikidata cross-check. A SPARQL query against Wikidata using P356 (DOI) and P5748 (retraction status) provides a third independent lookup. If any of the three registries flags a retraction, the reference is marked RETRACTED and the notice DOI is surfaced.
5. Compare. When the DOI resolves, the year, title, and journal you cited are compared to the registry record. Differences beyond a small tolerance are surfaced as MISMATCHwith a per-field delta — we don’t auto-rewrite your citation, we show you exactly where it diverged.
6. Sources.Every finding row includes a link to the Crossref, OpenAlex, and Wikidata records it drew from. The audit ships zero unsourced factual claims — if a claim needs backing and we can’t back it, we don’t make it.
No. Retraction Watch coverage inside Crossref is imperfect, and some retractions are indexed with a delay of days to weeks after the notice is issued. The audit is a strong sieve, not a proof of cleanliness. When we cannot confirm a retraction, we don’t claim one.
You get a MISMATCH row with the per-field delta (title / year / journal). We don’t rewrite the citation for you — the correct fix depends on your journal’s style, and blindly overwriting user copy is exactly the kind of quiet failure we’re trying to prevent.
PDFs are parsed client-side; only the extracted References text is sent to the server for reference-level lookups. Nothing about your unpublished results, figures, or Methods leaves your browser unless you explicitly paste it into the Methods lint on the paid tier.
Yes — the paid tier renders a print-ready PDF audit report with every source URL clickable. The report is designed to slot into the same submission-checklist field where journals ask for a manuscript-integrity statement.