The growing popularity of transmit-arrays (TAs) for various antenna applications is calling for effective analysis and optimization methods. TAs are usually electrically large, comprising thousands of unit-cells formed by subwavelength metallic scatterers. Full-wave optimization cycles needed to meet stringent specifications in terms of gain, cross polarization, bandwidth, scan-loss, etc. may be impaired by unrealistic required computational time and memory resources. To overcome this, we propose a modified homogenization method that unlike other approaches captures the internal reflections in the unit-cells and its resonances for each polarization, thus describing correctly unit-cells’ frequency response in the band of interest. We define equivalent dispersive anisotropic media for gradient TAs. These surrogate models enable fast analysis and optimization of TAs without compromising the accuracy. As an example, we analyze a TA composed of phase rotation (PR) unit-cells. PR unit-cells present wideband low axial ratio for a TA, but challenge the validation of existing homogenization methods. Detailed general description of the method is provided so that it can be applied to other unit-cells and avoid training time and resources required for machine learning based methods. Using the surrogate cells, the full-wave analysis time and memory of the TA reduces thirteen and four times, respectively.