This is a flaw with current AI in general. How can you trust anything it produces?

You either have to trust it implicitly, or verify the information manually, which you then have to weigh whether using the AI was worth it in the first place.

Depends on your use case and how precise and accurate the answer needs to be I suppose.