the problem with that is anything that can 'detect' an html element based on human-language will eventually fail to find something described with human-language due to crappy html on legacy systems.