Detect hallucinated references and score citation completeness — in seconds, for free
Paste any text or fetch it from a URL. The checker finds every in-text citation marker, parses your reference list field by field, flags entries that show hallucination risk signals, and gives you a score built entirely from what it actually finds — never a guess.
What this checks
Results will appear here after you run an analysis.
Deep Dive
A fabricated reference is built to look exactly like a real one — a plausible author, a real-sounding journal, a year that fits the timeline, sometimes even a DOI-shaped string. The structure is convincing precisely because it was generated to match the pattern of genuine citations. Reading a reference list line by line rarely catches this; checking each one field by field against what's actually verifiable does.
Every score starts at zero. Points are added only when evidence is found: a reference list exists, each entry has an author, a year, a title, a source, and an identifier, and the in-text markers correspond to that list. A document with no citations at all scores zero — not a misleading default. This means the score reflects what is actually present in your text, not an assumption about what a typical document looks like.
The checker cannot confirm a paper exists — that requires querying an academic database directly. What it can do reliably is flag the structural patterns most associated with fabrication: no identifier at all, a malformed DOI, an implausible date, a placeholder-style author name, or an exact duplicate elsewhere in the list. A CRITICAL flag means "verify this before you trust it," not "this is definitely fake."
Logic
The tool locates your reference list, splits it into individual entries, checks each one for five core fields, cross-references in-text markers against that list, and flags entries matching known hallucination risk patterns — entirely client-side, with no server round-trip for pasted text.
Methodology
Completeness, clarity, and reliability scores are each built from zero using a weighted composite of detected evidence; risk scoring per reference is additive across no-identifier, malformed-DOI, implausible-date, missing-author, placeholder-author, and duplicate signals, capped at 100 and tiered from LOW to CRITICAL.
References
Reference
Direct line
Need a workflow built for your specific stack? Describe the use case.