Free tool that checks whether each sentence of AI-generated text is actually backed by the original source text.
This tool is a standalone demo that lets anyone try out part of AnchorStrategy's core technology (citation fidelity checking) for free.
What this tool can and can't detect: This demo uses only the deterministic matching layer of the core technology (Layer A: exact-match checks on proper nouns / numeric tokens plus bigram similarity). It does not call an LLM for semantic judgment (Layer B). As a result, it cannot detect subtle misquotes that change a sentence's meaning without changing any proper noun or number — e.g. reordered clauses, flipped negation, or swapped cause/effect (a known limitation found during our own internal testing). Treat it as a tool for catching clear-cut cases: proper nouns or numbers that plainly don't match, or claim text that barely overlaps with the source at all.
It supports both Japanese and English input (company names, roles such as CEO/CFO/COO, and numeric matching, as well as sentence splitting, are supported for both languages). For English input specifically, generic role words such as "Director" or "Manager" are intentionally not tracked as role tokens: testing found they cause false positives ("target mix-up") whenever they merely appear as incidental context elsewhere in the source sentence (a known constraint).