crawler to start with Google search instead of seed page?

How can a crawler be designed to start by going to Google and entering keywords to find relevant pages to crawl... As opposed to starting with a seed page and only crawling the links from that page?