On-demand lecture videos play an important role in an increasingly online learning environment. However, there have not been many studies that conducted intervention experiments focusing on optimal methods of lecture video production. Most of these studies performed statistical analyses of indices such as video length, playback speed, and number of video player operations based on viewing records of existing lecture videos, such as massive open online courses. We therefore collaborated with an English lecture specialist to conduct an intervention experiment by creating two videos with the same topics, lecturers, and studios, but different editing policies. The results showed that for the video with clear chapter segmentation (but not the video without such segmentation), knowledge scores increased according to learning time. Detailed log analysis revealed that learners who viewed the lecture video with explicit chapter segmentation changed their behavior, such as seeking, changing playback speed, and skipping topics, depending on their own abilities. Based on the results, we propose that it is important for lecture video producers to organize chapters and design user interfaces based on the assumption that learners actively attempt to optimize their viewing behavior according to their abilities.