Trust · one of the 17 tools
Fact Checker
AI citation checker: every claim verified against its source
Fact Checker is the trust layer under every tool: claim-to-source verification against full text, retraction monitoring, an originality screen, and an audit trail you can hand to your examiner.
What Fact Checker does
Claim ↔ source verification
Every cited sentence is checked against the full text of the paper it cites, not its abstract, never the model's memory. No stored full text means honestly 'unverified', not a guess.
Four honest verdicts
Supported, partial, unsupported, and unverified are counted separately: a never-checked manuscript reads 'not yet verified', never '100%'.
Retraction is temporal
A paper can be retracted after you cite it, so it's re-checked at every consumption surface, including the final compile.
Originality screen
Verbatim-overlap detection against the literature runs on every compile. It catches copied phrasing before your institution's software does.
The audit trail
Every action and every check is logged; the AI-use disclosure document is generated from that real record, defensible to any committee.
Grounded by design
A reference that doesn't resolve to a real paper is rejected before you ever see it. That rule has no exceptions.
Who Fact Checker is for
Anyone using AI in academic work
Claims are checked against the cited paper's full text before you see them, not after a supervisor finds a problem.
Students who must declare AI use
A full audit trail and an AI-use disclosure document can be exported for an examiner or institution.
Supervisors and examiners
Verdicts are reported plainly: supported, partial, unsupported or unverified, never inflated.
How Fact Checker compares
- An AI detector
- Guesses whether text was AI-written, and false-positives honest students.
- Checking references by hand
- The correct method, and impossibly slow at thesis scale.
- Fact Checker
- Verification runs before output reaches you, and every check is recorded rather than asserted.
Related guides
Why AI makes up citations, and the fix
ChatGPT and other chatbots invent references because they predict text, not truth. The mechanism, the five failure modes, and the structural fix.
How verified citations work: the four verdicts
Supported, partial, unsupported, unverified: what the Fact Checker actually checks, why it never grounds on abstracts, and how to fix each verdict.
How to check if a paper has been retracted
How to check any paper for retraction before you cite it, how to sweep a full thesis bibliography, and what to do when a source is withdrawn.
Is there an AI that doesn't make up references?
Yes. Fabricated references end when citations are database objects verified against full text. The by-construction answer, plus a test checklist.
Fact Checker questions
Is there an AI that does not make up references?
Yes. Citations here are database objects that must resolve to real papers, and every AI-written claim is checked against the cited paper's full text before you see it.
How do I check whether a citation is real before submitting?
Every claim carries a verdict: supported, partial, unsupported or unverified. Nothing is inflated, and unsupported claims are flagged rather than hidden.
Can I prove to an examiner that my citations are genuine?
Yes. A full audit trail and an AI-use disclosure document can be exported for a supervisor, examiner or institution.