One cause for this issue might be the incorrect usage of Linear Analysis Points. Please keep in mind that using "Open-loop Input/Output" will break existing feedback loops. If you are interested in performing linear analysis on a closed loop model, a combination of "Input Perturbation" and "Output Measurement" points should be used to designate the input and the output, respectively.
It is also important to keep in mind that linearizing a non-linear model will inherently cause the step response to look different. For instance, if your model contains a saturation block, it will be converted to a gain during linearization. As a result, if the signal is reaching the saturation limit during simulation, this behavior cannot be reflected in the linearized step response.