We're talking about LLM-driven search engines here, the assumption is that they will always need up-to-date information. A "strong LLM" can't give you to latest on the presidential election if its knowledge cut-off is in 2023, so these companies "solution" is to scrape today's New York Times and get the LLM to write a summary.
(In fact, there's value in trying to filter excess crap out of existing training sets.)