GLM Link function and model function

Hi,

For a quite while, i'm trying to make sense of GLM.

mdl =  fitglm(X,y,'linear','Distribution','poisson')

I saw that example on the MATLAB website. The question is, aren't we using log function to predict the data while using poisson distribution with Glm ? If not, is link function only used for converting predictors to a given distribution μ. How does this help us to predict the data ?