A system and method for implementing keyword prediction for search engine marketing is presented. Historical search data is analyzed to find top search patterns in the form of n-grams. Keywords from historical search data are analyzed to generate n-grams for various values of n. The n-grams can represent trending search term patterns, as opposed to merely finding trending keywords. Thereafter, each of the n-grams is associated with one or more categories and/or sub-categories. The categories and sub-categories are thus associated with one or more n-grams. When a new product is introduced, the new product can be associated with n-grams based on the categories and sub-categories associated with the new product. Advertisements can be bid upon based on the associations. Other embodiments also are disclosed.