My AI chatbot kept apologizing for existing, so I rebuilt its prompts

Last month I was testing a customer support bot for my small shop in Austin and it kept saying sorry for literally everything, even when someone typed 'thanks'. I spent a whole evening tracing the training data and found my example logs were full of overly polite responses I'd cherry-picked from old emails. I rewrote the prompt to include blunt, helpful phrasing and added a rule to only apologize when there's an actual mistake. Now it tells users 'that'll be $12.50' instead of 'I deeply regret any inconvenience regarding pricing.' Has anyone else hit a bot that overcorrects into groveling mode?

0 comments

0 Comments

No comments yet

Be the first to share your thoughts on this discussion.