“A bias towards verbosity in chatbots may arise from the labeling process, where raters evaluate single messages in isolation and prefer a comprehensive response over a shorter, more interactive one.”