They're useful for embeddings, which let you turn articles (and images and other content) into a huge array of floating point numbers that capture the semantics of the content. Then you can use a vector database to find similar items to each other - or similar items to the user's own search query.

I wrote a big tutorial about embeddings a couple of years ago which still holds up today: https://simonwillison.net/2023/Oct/23/embeddings/