In this paper, we present a novel approach for the design of rate-maximizing channel shortening detectors with soft feedback side information for frequency-selective channels. The detector is a soft-input soft-output detector and constitutes one of the components of an iterative receiver. The design optimization is performed from an information-theoretic perspective where we maximize the achievable rate during each step of the iterative process. Our proposed detector consists of a front-end filter whose coefficients are given in closed form in addition to a convex optimization procedure which provides the branch labels of the trellis and the feedback filter coefficients. The detector can be implemented as a BCJR-type algorithm operating on a trellis where the number of states is a user-defined parameter.