I wonder if one could circumvent this by making their initial response appear to come from another spammer to get 2 spammers talking to each other.

It would also avoid problems with canned responses failing the Turing test.

Spam filters could do very little, because it's almost impossible to tell whether a conversation was run by a bot - and it's outright impossible to do so for a single response.

However, you raise an interesting point: blacklists could reasonably be put into place. I'll make a merge request to use Spintax.