LLMs do not return truth. They return answer-shaped text that is, sometimes, correct.

Unless of course someone configures them to return text that deliberately hides the truth.

https://newrepublic.com/post/216016/trump-rigs-americagov-ai-chatbot-fact-check-lies