Even when AI is given search results it can struggle with anything that’s not simple. Kagi has a search integration and it references the websites, even then I often find mistakes, although much less without it.

I don’t trust LLMs without this integration

Considering in much of the world Google defaults to AI answers now, it’s just a matter of who makes up less incorrect shit

they released a search engine where the model reads the first link before trying to answer your request