Skip to main content

The AI answered something wrong, or made something up. What do I do?

The real fix is almost always what the AI knows or how strictly it is told to stick to your services, not a bug to report.

Your AI told a customer something wrong, out of date, or that was not true at all. There is no "un-say that" button, because there is nothing to undo, the AI answers fresh from what it currently knows and the rules you have set, so fixing what it knows or how it is allowed to answer is the actual fix, not a bug report.

Work out which kind of wrong it was

  1. It got a fact wrong, a price, an hour, a policy. This is a missing or outdated knowledge base entry, almost always. Check What are knowledge gaps, and how do I fix them? first, it shows you the exact questions the AI could not confidently answer in the last 30 days. If the wrong answer is not in that list, correct or add the entry yourself, see How do I add something to my knowledge base by hand? The AI reads the correct answer the very next time it is asked, there is nothing further to save or publish.
  2. It answered something entirely outside what you offer. Check how strict your AI is set to be, under Settings, then AI Assistant, Topics & Rules. Open mode is allowed to use its own general knowledge for anything adjacent to your business, which is exactly where a plausible-sounding but wrong answer can slip in. If that is not what you want, switch to Balanced or Strict, see How do I set the AI's tone and what it can and can't say?
  3. It said something it should never say. Never-say rules exist for exactly this. If it broke one of your own rules, or one of Peakify's platform-wide safety rules, that is worth reporting to us directly rather than only fixing your settings, see below.

What we do on our side, and what we do not

Depending on your account, a second automatic pass checks a drafted reply against your rules before it is sent, and swaps in your off-topic response and brings in a human if it catches a violation. This is a rules check, not a fact checker, it cannot tell true from false, only whether an answer strayed from what you have told it not to talk about. It will not catch a wrong price or a wrong hour, only your knowledge base does that.

If a rule was genuinely broken, not just a knowledge gap

If the AI said something that ignored a never-say rule you had set, or said something unsafe, that is not something you can self-serve fix by editing your settings, since the rule was already there. Open Help in your sidebar and email us the conversation, so we can look at why the rule did not hold.

Was this useful?

More in Something is not working

The other guides people read alongside this one.

Did this not cover it?

Start a chat and the assistant will try first. If it cannot help, it hands you to a person.

Or email support@peakifyhub.com. We answer within 1 business day, Mon–Fri.