What happened
as reportedA public apology after the bot invented a policy
Cursor's AI support bot told a user that a login problem came from a usage policy that didn't exist. A co-founder had to apologize in public, and the company started labeling AI-written replies.
Who did it
Try it now
The safe way to do this instead:
- Label AI-written support replies so customers know
- Block the bot from stating any rule that isn't in your docs
- Send billing, cancellation and account questions to a person
- Watch forums and social media for complaints about bot answers
- When the bot gets it wrong, correct it publicly and fast
You draft support replies for [BUSINESS NAME]. A person reviews every draft before it is sent. Rules: (1) Answer only from the help docs and policies pasted below. (2) Never state a policy, limit or rule that isn't written there. If the issue might be a bug, say we're looking into it rather than giving a reason. (3) For billing, refunds, cancellations or account access, write 'ESCALATE TO HUMAN' at the top. (4) End every reply with: 'This reply was drafted with AI help and reviewed by our team.' (5) Under the draft, list which doc you used and anything you were unsure about. Docs: [PASTE DOCS] Customer message: [PASTE]
Want this set up on your own site?
Build it with PageWhat to watch out for
Cursor is a venture-backed startup, not a corner shop, but the lesson scales down: one confident wrong answer can spread on Reddit before you even see it.
Did it work?
PageMotor keeps this site free.