There are legitimate reasons to do that, of course. Maybe I am trying to find info about some niche topic or how to do X, I ask an llm, the llm goes through some search results, a lot of which is search engine optimised crap, finds the relevant info and answers my question.
But if I wrote articles in a news site, I am supported by ads or subscriptions and see my visits plummel because people, who would usually google about topic X and then visit my website that I wrote about X, were now reading the google summary that appeared when googling about topic X, based on my article, maybe I would have less motivation to continue writing.
The only end result possible in such a scenario is that everything commercial of some quality being heavily paywalled, some tiny amount of free and open small web, and a huge amount of AI generated slop, because the value of an article in the open internet is now so low that only AI can produce it (economically, time-wise) efficiently enough.
In 3, although you do not specify, I assume you mean that a bot requests the page, as opposed to you visiting the page like in scenario 2 and then an LLM processes the downloaded data (similarly to an adblocker). It is the former case that is a problem, the latter case is much harder to stop and there is much less reason to stop it.
This is the distinction: is a human present at the time of request.