It has various categories of replies, which right now you can specify when you forward a message (it sends you a management URL back). Later on, when I have enough data, I'll do some simple ML to have it pick a most likely category as a default.
Each category comes with its own set of responses, but you can see where the category changed because the bot will start talking about a product and then switch to talking about funds.
It'd be more effective if it didn't re-use responses later in the chain (unless it'd totally run out of alternatives). Although, arguably, it's funnier when the spammers/scammers do continue to find new ways to reply to the same couple of questions.
Each category comes with its own set of responses, but you can see where the category changed because the bot will start talking about a product and then switch to talking about funds.
The code is here if you want to see:
https://gitlab.com/stavros/Spamnesty/