Each of these systems are probably different, but the way one vendor's system has been explained to me is that an automated classifier will eventually visit most sites and will come up with a computed category for only the main page. The system does not want to give full access to the site with a lot more potential harm, so if a customer visits one of these automatically classified sites and sees it is restricted, they can submit a classification at which point an employee will verify the classification and unlock the full site. Sometimes the restrictions are bizarre, such as allowing the HTML content to be viewed, but not the CSS. This process takes up to 24 hrs. From what I've seen, these automated classifiers are not always friendly to hostnames tacked onto a domain name (e.g., fsf.org/donate would probably be better than donate.fsf.org). They also routinely prevent visiting new domain names thrown up in Show HN posts until the URL is submitted for classification and gets whitelisted.
