Top of the page, right hand column. With an image called, "btn_donate_LG.gif." A simple check of the text of the page would reveal that it does, in fact, say donate. It says it inside the name of an image, but if you're searching the page source that doesn't matter: run a search for donate and you get a hit.
It shouldn't be hard to make it so that the "I'd totally donate if you just had a button," (paraphrase) spam didn't show up on blogs with a donate button. Just have a bit of code that makes it search the page's source for the word "donate" and if it gets a hit use some other flavor of spam on that particular page. It's all still automated, it still requires no human oversight, but it makes it marginally more likely that someone won't be able to tell in one second or less that the message is definitely spam.