Unsupported Claim Detector

Paste an AI answer and its source documents. Get every unsupported claim — with the reason it fails and the evidence that would fix it.

An unsupported claim is any statement in an AI output that its source material does not establish. Finding them by hand means re-reading every answer against every source — which is why most slip through. MissingProof automates the check at the level where fabrication actually happens: individual claims and the reasoning links between them.

What "unsupported" actually means — three different situations

Detection tools that output one score treat these as the same problem. They are not:

• Missing fact. The claim asserts something no passage establishes. The data may exist in a document you didn't provide — or nowhere.
• Fabricated reasoning. Every underlying fact checks out, but the claim connects them with a relation — because, therefore, concluded, will — that no source states and the facts don't entail. This is the failure aggregate scores systematically launder, because the claim is built from genuinely supported text.
• Contradiction. Your sources state the opposite. Crucially, this one is unfixable by adding evidence — the claim itself must change.

A worked example

Notice what a similarity score would do with this example: the answer overlaps almost completely with the sources, so it scores as faithful. The fabrication is invisible at answer level and obvious at claim level.

What you get per detection

1. The claim, isolated and quoted
2. Its verdict, with the supporting evidence span for anything that passes
3. The failure locus — PREMISE, BRIDGE, or CONTRADICTION
4. Whether the gap is fixable, and if so the exact statement your sources would need to contain
5. Where that statement plausibly lives and who typically owns such a document