I've been looking at a lot of spam lately as I'm developing services based on text classification as a web service at classifyr.com, and spam is kind of the obvious test. Initially, the site was getting 8200 spam attempts a day with a captcha that was 99.9% effective. The classifier is 99.99% effective and has cut down the number of spam attempts to almost nothing.
The ones that still get through on rare occasions copy legitimate content and usually link to sites that aren't inherently spammy (like grey-market pharmacies) for SEO purposes. I suspect these are actually posted by humans; a post can classify as suspect rather than spam or ham and these always do when they get through. When that happens, the user is asked a trivia question related to the site's subject matter. I think it extremely unlikely a spam bot would be able to answer correctly.
The ones that still get through on rare occasions copy legitimate content and usually link to sites that aren't inherently spammy (like grey-market pharmacies) for SEO purposes. I suspect these are actually posted by humans; a post can classify as suspect rather than spam or ham and these always do when they get through. When that happens, the user is asked a trivia question related to the site's subject matter. I think it extremely unlikely a spam bot would be able to answer correctly.