Abstarct: Generally, automatic keyword extraction methods attempt to get better results in criterions such as recall and precision, which show the ability of the method to perform the desired action as an author. According to the undeniable role of search engines in today's world, it seems that keywords selection methods should pay the same attention to web content accessibility by search engines and the information retrieval criterions. In this thesis, a new Automatic keyword extraction method is presented which acquires as good recall/precision as other methods and at the same time increases the contents accessibility by search engines. To extract keywords two score functions are used. The firsrt function is a statistical keyword selector and the second one is a search engine results evaluator. The keyword selector function attempts to get a better recall/precision, while the search engine result evaluator function optimizes the first one, using genetic algorithm and search engine results during training phase. The introduced keyword extraction method also uses a new post-processing pahse which employs attention attraction strings to improve precision without losing noticeable recall. Experimental results show that using proper documents in training phase, reasonable balance of the recall, precision and web content accessibility can be obtained