Those who are harmed decide. 230 is about protecting companies from law suits filed by users.
The whole "end of free speech" issue comes not so much from the government sensor really (that's still firmly restricted by the first amendment) but from companies themselves banning any content or accounts that might get them sued.
But if that risk is limited only to what they recommend outside a user's direct boolean search and filters, they can still host content without concern. But they need to be sure they know and approve exactly what their algorithms are pushing onto people.