I'm waiting for a day there will be a capable LLM agent that would be able to crawl hundreds of results, read them all (working around all the obstacles, not spewing "reading content failed"), filter the marketing noise/SEO copypasta bullshit, find the meaningful bits, and present a nice summary.
Current LLMs can go crazy just by seeing a single well-chosen word, see https://www.lesswrong.com/posts/aPeJE8bSo6rAFoLqg/solidgoldm... for an amusing exanple. I think we need some breakthrough in adversarial training before LLMs can survive the very adversarial environment of SEO.
