I don't think you can say many useful things. E.g., if you conider an embedding of a Levi $M \subset GL(2)$, then the pullback should correspond to restriction. Cuspidal automorphic representation or Eisensteries representations are infinite-dimensional, but $M$ automorphic representations are one-dimensional. On the other hand, there are positive examples where useful things can be said, e.g. Labesse-Langlands for $SL(2) \subset GL(2)$ .
Also this seems relevant: Restriction map between spaces of automorphic formsRestriction map between spaces of automorphic forms