And how, exactly, do you "verify" that data?

If we had an automatic way to distinguish true from false statements, we'd have AI already.

The big language models are training on terabytes of information - the idea of paying people to go through this and mark it as true or false is a nonstarter.