Analyzer · Citations
Fabricated references, exposed before a reviewer finds them
Every reference is verified against the registries that actually issue them — DOIs resolved, papers matched, URLs probed. In-text citations are cross-checked against the bibliography both ways. AI-fabricated citations are the fastest way to lose a reader's trust; this catches them first.
Real-registry verification
DOIs resolve through content negotiation — not a single database — so DataCite and arXiv works verify correctly, not just journal DOIs.
Both-ways cross-check
In-text citations without a bibliography entry, and bibliography entries never cited in the text, are both surfaced.
Four verdict levels
Verified, partially verified, could-not-verify, and suspicious — with the evidence for each, so a flaky DOI isn't treated like a fabricated paper.
Uncited-claims scan
Specific factual claims with no citation nearby are listed — the places a reviewer will write "citation needed".
How it works
01
Include the reference list
Paste the document with its bibliography — the cross-check needs both sides.
02
Verify
References are extracted verbatim and checked against DOI registries, CrossRef, arXiv, and live URLs.
03
Fix before submission
Suspicious and unverifiable references are listed with what was checked — fix or replace them before anyone else looks.
Verification, not endorsement
Verified means the reference exists and matches — not that it supports your argument. Could-not-verify often means a source that's real but offline or paywalled; suspicious means the registry contradicts the citation. The evidence for each verdict is shown.
Straight answers
Can it catch AI-invented citations?
That's its main job. Fabricated references usually have DOIs that don't resolve, authors that don't match, or papers that don't exist in any registry — exactly what registry verification exposes.
Which citation styles work?
Extraction is verbatim and style-tolerant — APA, MLA, Chicago, and imperfect mixtures all parse. Broken formatting affects confidence, not whether checking happens.
Why did a real paper come back could-not-verify?
Usually a stale URL or a registry hiccup. Could-not-verify is deliberately separate from suspicious — it means we couldn't confirm, not that it's fake.
The other instruments
Free to start — 50 credits a month, no card.
Citations is one of six instruments sharing one document — check, verify, and polish without pasting twice.