Usability testing is not a singular phenomenon. In other words, you don't test your product (or item or document, or whatever) and then improve it according to the test results, and then quit.

True usability testing takes place in an iterative fashion. Once a usability test is completed, the results should be used to modify the product accordingly, and then more tests should be run. This constant usability testing is like sanding a piece of wood to a glass-like finish. The wood starts out rough, but then becomes smoother and smoother each time a new, finer grit of sand paper is used.

And, like sanding, the improvements at the beginning are easily seen. The product becomes quickly improved  and troubled areas--aspects that most all users pointed out as problems with the product--are resolved. But as more and more tests are run, problems are found and resolved, the problems become smaller and more individual (that is, not nearly as many people agree that they are, in fact, problems).

This process happens every time you utilize a search engine. Searches are tracked and recorded, as are the results of those searches. The people behind the search engine can tell if a search was successful or not depending on whether the user selected one of the choices offered or did something else. Hence, user data is constantly compiled to help them provide a better product (to see Google's discussion about user testing, especially iterative testing, select here).