When I ask Grok about a video attached to an X post, how much of the clip can it actually inspect? I’m wondering whether it analyzes the video frames themselves, uses an available transcript or caption, or mostly answers from the post text and surrounding context.

If the clip is long, does adding a timestamp or describing the moment I mean make a meaningful difference?