I was just thinking that. You'd only have to beat out all the overly-literal bots on a single day, so I'd imagine the complexity of the problem is dramatically reduced. I think ideally you would train a bot to recognize some measure of satire/joking, yet you would only want to execute on the press releases/articles that were just on the very edge of satire, so as to detect satire that fools as many general purpose bots as possible, as fast as possible. I doubt there is really that much of this happening, relative to the size of the market, though.