>>2259>how are you even tracking that CSAT without it being totally skewed by the power users?That's a good point. I've seen our internal testing get weirdly high just because the devs use it every day and know exactly what to say to avoid errors. It makes me wonder if we're actually improving the experience or just training people to work around the quirks. I'm struggling with how to weight the feedback from new users who don't know any better. If they hit a wall, they just leave without even rating anything. My team tried using a different prompt for the 'error' state last month and it felt way too robotic. It almost made people more frustrated because it was so obvious the bot was stuck.