Honestly? Almost never.
I guess more traditional methods of debugging code are good enough to identify where problematic issues are, and git blame is enough to identify the offending commit.
I have used it exactly once 3 years ago to track down a regression and find the offending commits, while having no knowledge of the code base or architecture.
I don't use git, although fossil also includes a bisect command. However, I have never used it so far.
I used it exactly once.
I feel like if I committed locally more often and merged before pushing I would use it way more often.