Computational linguistics

A next-token model assigns P(token | prefix). If the prefix changes while the token stays the same, what is generally true?