Right, but the "platform immunity" thing applies equally to all platforms. You're instead proposing to extend immunity to some platforms and not others based on how they choose to recommend things. If you (like I do) believe that recommendation is a form of speech, this creates government favored classes of speech.
To word it differently: "Platforms that recommend/aggregate speech in government sanctioned ways are protected from liability, while those that do not are not". That's, while not open and shut perhaps, certainly spooky.
The law already does something like that. If I run the hypothetical blog host conspiracy-blogs.com, and I manually select ten posts I like to put on the front page every day, I can be prosecuted if I select a post for that saying something like:
Tomorrow at 10 AM, join me on the steps of the capitol. Bring your gun and a sign that says "we will shoot you unless you vote no on bill 123.".
The action described is a crime, soliciting others to participate in it is a crime, and my deliberate choice to show it to a larger audience is likely enough to subject me to accomplice liability that virtually any criminal lawyer would recommend against it. If violence actually occurred, there would be an ever lower bar for the victims to clear to sue me. Planning crimes is unambiguously not protected speech.
The proposal described would subject any sufficiently editorial algorithm to similar liability.
To word it differently: "Platforms that recommend/aggregate speech in government sanctioned ways are protected from liability, while those that do not are not". That's, while not open and shut perhaps, certainly spooky.