Predicting Performance of Responsive Search Ad Surface Texts

Abstract

Despite the popularity of responsive ads that dynamically combine text assets to best suit individual search queries, there is a distinct need for predicting how the final displayed text might perform. In order to serve those needs, we are building a model that leverages the impressions of responsive search ad permutations and aggregate CTRs of the responsive ads they belong to. This paper describes the thinking and assumptions behind this system as well as some of the challenges involved.