We have the following problem: We have deposited a website as a KB. We want the bot to answer the user's questions exactly from the content of the website. We also told the bot to always provide the link where it found the information on the website. Unfortunately, it is now the case that the bot invents some answers and even the links for them, which then do not exist at all. We've also said in the personality settings that he's forbidden from lying. But that's no use. How can this be turned off?

Seems weird, from my personal knowledge the answers from the KB are straight-up picked from the source. Are you using an AITask somewhere in the process?

Yes. we use the AITask. does this have anything to do with the fact that we use the web search option as KB (but we use the optin specific page)?

hi @rapid-agency-63781 👋 the AI Task allows for the possibility of hallucinations, whereas the KB (even when web search specific page option selected) should not have hallucinations. A few troubleshooting Qs: are you feeding the KB response into the AI Task? do you have access to the same content as content you can upload?

hi @acoustic-hair-60678 - thanks. I think I have found the problem. Can it be that the AI also translates links. We have built our bot in German and the website we use for the KB is also in German. If we now ask the bot in German where to find what on the website, it probably translates the question into English searches in the KB, finds the answer and translates the link from German to English for processing and then back to German to give this to the user as an answer. Only apparently the translation is inaccurate because there can be different matching words in German and English. Confusing - but can this be possible? If so, how do we turn this off?

Following, had the same issue