LLMs are already grounding their results in Google searches with citations. They have been doing that for a year already. Optional with all the big models from OpenAI, Google, xAI
And yet they still hallucinate and offer dead links. I've gotten wrong answers to simple historical event and people questions with sources that are entirely fabricated and referencing a dead link to an irrelevant site. Google results don't do that. This is why I use LLM's to help me come up with better searches that I perform and tune myself. That's valuable, the wordsmithing they can do given their solid word and word part statistics.
People talk about LLM hallucinations as if they're a new problem, but content mill blog posts existed 15 years ago, and they read like LLM bullshit back then, and they still exist. Clicking through to Google search results typically results in lower-quality information than just asking Gemini 2.5 pro. (which can give you the same links formatted in a more legible fashion if you need to verify.)
What people call "AI slop" existed before AI and AI where I control the prompt is getting to be better than what you will find on those sorts of websites.
And even if we solve this problem of hallucination, the ai agents still need a platform to do search.
If I was Google I’d simply cut off public api access to the search engine.