Recommendations in the form of ads targeted to humans is a perfect usecase for LLMs. There is no right answer, and the interpreter is a human. Hallucinations dont matter, and if the targeting is a tad bit better, that justifies the investment.

Am I supposed to ask for permission when I talk to my friends about how attractive I think people of the opposite sex are?

Many social media users are gullible enough to be convinced to act irrationally.

A well-trained AI on great data (which Facebook has) is pretty good at sorting signal from noise, i.e. advertisements that would or would not appeal to a particular user.

This post is on to something.