Free tool

Do your references actually resolve?

AI writing assistants invent citations that look perfect — a plausible title, real-sounding authors, a well-formed DOI that points to nothing, or to an entirely different paper. Paste your reference list and this tool resolves each cited DOI and PMID (up to 25 per run) against the public registries, so a non-resolving or mismatched identifier surfaces before a reviewer or an editor’s checker finds it.

§01 The check

Paste your reference listyour manuscript stays in your browser

Paste your bibliography, or just a list of DOIs and PMIDs. Up to 25 identifiers are checked per run. Only the identifiers — and, for a recognized reference list, the single reference line each one sits in (used to check it resolves to the right work) — are sent to RigorMD; the public registries receive nothing but the bare identifier. Nothing is stored.

§02 What it checks — and what a ✓ does not mean

Each identifier is resolved by bare identifier only against three public sources — Crossref, the DOI handle registry (the DOI system of record, which covers registration agencies beyond Crossref), and NCBI PubMed. Three outcomes matter:

Does not resolve — the DOI or PMID is not found in the registries. The most common cause in 2026 is a citation invented or garbled by an AI assistant, but publisher indexing can also lag for very recent papers — so verify each one before you assume the worst.
Resolves to a different work— the identifier is real, but the registry record’s title, year, or first author clearly doesn’t match the citation as printed. That is the signature of a real DOI pasted onto the wrong reference.
Resolves — the record exists and is consistent with your citation. This confirms the identifier, not the science: a ✓ does notmean the cited work actually supports the claim it’s attached to, and it does not check retraction status, journal quality, or whether the reference is the best one for your point.

When a registry is briefly unreachable, that identifier is listed as not checked — never assumed to resolve, and never counted as a problem. This is reference-integrity QC. It is not a plagiarism check, a fabricated-data detector, or an accusation of misconduct — it reports what the registries say, with the date, so you can verify every line yourself.

§03 Private by construction

Your manuscript never leaves your browser. The identifiers are extracted locally, and only the bare DOIs and PMIDs — plus, when you paste a recognized reference list, the single reference line each one sits in, used solely to check the record matches — are sent to our server to run the lookups (up to 25 per run). The public registries themselves receive nothing but the bare identifier. Nothing you paste is stored. This is the same reference layer that runs inside a full RigorMD review, exposed here for free.

§04 From citations to the whole manuscript

Broken references are one thing reviewers catch. The $25 RigorMD pre-submission review checks the rest: it recomputes the statistics it can rebuild from your reported numbers, appraises the methods across two independent engines, and resolves your references as one layer of a severity-scored, quote-grounded report — the same reference check you just ran, plus everything around it.

Start a full review · $25 See every check we run

A check, not a certification. This tool reports whether each cited identifier resolves in the public registries as of today — it does not certify a reference list, detect plagiarism or data fabrication, or assess whether a source supports your claim. Registry indexing can lag for recent work; verify every flagged reference against its source before acting. RigorMD flags; it never certifies.