Architecture behind - how AI reads the internet ?

You type a complex, multi-part question into the search bar: “What were the key takeaways from the Federal Reserve meeting yesterday, and how did tech stocks react?”

Instantly, you see three or four “Sources” pop in at the top of the screen. A fraction of a second later, a coherent, beautifully formatted answer begins streaming across your screen, peppered with inline citations like [1] and [2] that perfectly map to the source links.

It feels like the AI just quickly read the internet and wrote you a report. But what is actually happening behind the scenes in those crucial three seconds?