What if the AI writes something wrong?

In Review Mode nothing reaches a lead without your approval — you edit the draft before it can send, and your edits are fed back into how Monty writes, so it steers away from the phrasing you corrected (rejecting a draft stops the send, but it teaches Monty less than an edit does). In Autopilot, guardrails bound what Monty may say: every price, link, and plan name in a draft has to match a fact you've given us, guarantee and certification claims you haven't approved are escalated to you instead of sent, and so is anything else risky or unusual. Key state changes are recorded in the lead's history so you can trace what happened. If something still slips through, tell support — reply-quality reports go straight to the people who tune the drafting engine.