Your argument is that of a straw man, it does not actually address the issue at hand. They are not composing a message in the way that generative algorithms like those we have now can, out of whole cloth. I would also argue that one could not copyright the autocorrect strings on iOS, see below for an example:

“Okay dear I just don’t think you know how to do anything for me to be able and to be honest with you lol”

This is just me selecting words that apple has determine are likely to be used together. There is no agency to it, and I would argue that, given the legal precedents at play, are not copyright eligible. Grammarly for example recognizes grammatical patterns, based on rules of language that are defined (as much as any language can be said to have rules) and suggests more technically correct ways to say what is being supplied by the author. There is still an author, however, with editorial control over the content, which they generate out of whole cloth and supply to the algorithm