Present day action logics usually depend on a notion of goal-orientation to identify rational agents. This notion might become misleading, because goals are typically context-dependent, whereas goal-statements are typically made independent of the context. Supplanting goals by preferred possible worlds can overcome this problem, if the underlying preference relation is properly identified. The paper distinguishes four kinds of preference relations, and provides both syntactic and semantic characterizations for each of them, relying on the notion of minimal change from conditional logic.