Personalized advertising is widely believed to be an effective persuasion strategy. A typical personalized advertising process consists of two phases: The message sender first “learns” the message receiver’s preferences, and then “matches” the message to that person according to his or her preferences. The present study argues that this process may be problematic because it assumes that an individual’s preferences are always stable (i.e., preferences remain the same over time) and extreme (i.e., preferences are highly polarized). Through a 2 (message type: personalized vs. nonpersonalized) × 2 (preference stability: high vs. low) × 2 (preference extremity: high vs. low) between-participants experiment ( N = 227), it is shown that the effectiveness of personalized advertising is moderated by preference stability and extremity. A new conceptualization of personalization is proposed based on the study results, and how the two phases of personalized advertising may be refined is highlighted.