Screen submissions for fabricated, retracted, or hijacked references — at scale.

Hallucite checks every citation in a submitted manuscript against CrossRef, PubMed, Semantic Scholar, and OpenAlex — producing auditable evidence your editorial team can rely on.

What a quick existence check misses

The most common way a fabricated citation hides is not a made-up title — it is a real DOI that resolves to a completely different paper. A does-the-title-exist check passes it; Hallucite's identifier integrity check does not.

Existence check only

DOI 10.xxxx/example → title found in CrossRef. ✓ Passes.

Verdict: no problem detected

Hallucite identifier check

DOI 10.xxxx/example → resolves to a different paper. Cited title, authors, and year do not match the record at that identifier.

Verdict: Discrepancy — identifier mismatch

Hallucite checks facts, not authorship. Every identifier is resolved live against CrossRef, Semantic Scholar, OpenAlex, and PubMed.

What we check — and what we don't

What Hallucite checks

  • Existence: the cited work is in an indexed database.
  • Identifier integrity: the DOI or PMID resolves to the right work.
  • Metadata consistency: authors, year, journal and title match.
  • Retraction status: the work hasn't been withdrawn.

What we don't do

  • We don't score whether your text was written by AI.
  • We don't yet judge whether a source supports the specific claim it's attached to.

Built for editorial workflows

  • 1

    Auditable evidence per reference

    Every reference in a submission receives a structured verdict — Verified, Discrepancy, Not Found, or Retracted — with the exact fields that do not match. The evidence is exportable and reproducible.

  • 2

    Batch processing and API access

    Screen hundreds of manuscripts in parallel via the REST API, or upload reference lists directly. Results are returned in JSON or CSV, ready to feed into your existing editorial management system.

  • 3

    Exports for your editorial record

    Download structured reports in PDF or CSV format for each submission. Attach them to the editorial record, share them with reviewers, or archive them as part of your post-publication audit trail.

Measured on the 100 k-case synthetic benchmark

92%Precisionbenchmark
79%Recallbenchmark
85%F1 Scorebenchmark
65%DOI-hijack detectionbenchmark
See our accuracy →

Integrate reference screening into your submission pipeline

Contact our team to discuss volume pricing, API access, and integration with your editorial management system.