My chatbot quoted a customer a fake $40 refund and I had to fix it live
I was demoing our support bot to a client in Austin and someone typed a refund question, so it just made up a policy and promised $40 back. I jumped in, killed the reply, and apologized, then spent that night adding a hard block so it can only answer from our own help docs. How are you all catching these made-up answers before a customer sees them?