Large introductory or general-education geoscience classes present opportunities to educate a broader group of students about societally relevant geoscience topics, as well as potentially recruit geoscience majors. Personalization is a well-established driver of student engagement, but large classes present challenges in creating connections between individual student interests and class topics. Advances in generative AI could make personalization more feasible. In Fall 2025, students in a large general-education geoscience class at University of California, Irvine were asked to watch a ~4-6 minute video each week connecting a topic from in-person lectures to a broader societal context. The scripts for the videos were generated using OpenAI's LLM GPT-4o and then were edited by the instructor as necessary for both accuracy and style preference. The scripts were then turned into videos by compositing and editing together an AI avatar of the instructor, bullet point summaries, images, titles, credits, and music. In alternate weeks, students were assigned through Canvas either a 1) non-personalized AI video, 2) a choice of three AI videos approaching the topic from a scientific, business, or policy perspective, or 3) a human-recorded version of the non-personalized AI video. To encourage students to watch the videos, two questions from a 12-question weekly online quiz were related to the video content. At the end of the quarter, students were surveyed to gather feedback regarding the different video formats. The data from the 322 respondents in the class shows that students have a clear preference for human-recorded videos over AI videos, but also showed a clear preference for personalized AI videos over non-personalized AI videos. We will also show results from the student survey that highlight the main reasons why students ranked the videos in this order, and examine whether student major influenced responses.