Context personalization, or matching instructional tasks to students’ personal interests, has been found to be a beneficial approach within digital learning platforms. Advances in large language models allow for personalized tasks to be generated with increased efficiency. While best practices for using generative AI in K-12 settings accentuate the importance of a human-in-the-loop, little is known about how teachers can use AI to personalize instructional tasks for their students. In the present study, we explore 7th grade teachers using ChatGPT to create personalized versions of mathematics problems, which were subsequently assigned to 348 students in the ASSISTments platform. We found that teachers’ efficiency for personalization varied widely, and that they often used prompts to swap out contextual or popular culture references in problems or to correct contextual inaccuracies in the problems ChatGPT posed. Further, we found that students may respond more positively to problems that teachers spent more time personalizing with ChatGPT, and that students were sometimes attuned to different personalized elements of problems than teachers.