You don't, but that's the wrong question. How do you know they're accurate?

I feel it's not that controversial to treat LLM output as unreliable without proper verification. So even if the output were accurate, if it's LLM-generated without proper verification I don't trust it.