In my experience LLM chatbots are a net improvement because they understand "put me in relation with a human" as opposed to the scripted ones that only spit out the FAQ.
I guess that an LLM is harder to scope and retain, and does what it is instructed to. While traditional chatbots can only do what they are programmed for (e.g. follow a scenario), LLMs are easier to sidetrack. Thus you as a client may have an easier time escaping their context, especially when it's a generic LLM integration done by underpaid contractors.
It really depends, but at least here in france many companies start having support over whatsapp/messenger/the likes. They used to suck hard a few years ago, but my recent experience with sncf connect (french railway company) was surprisingly good given my issue was working around an idiosyncracy of their system.
YMMV as we are talking about chatbots and not plain old forms, thus the interactivity is better and feedback loop to escape the LLM's context is faster.