It is really great to see this researched and documented. While not every A/B test is proof, it is compelling in this case - and confirms my long-head view that these chat bubbles are just bad UX. There is a lot of other recent "UX best practices" that are actually worst practices, but because every unicorn is doing it, teams often feel they should do the same - without asking whether this kind of "engagement" is really serving the core product management objectives.