What happens when the chatbot asks the team

22 October 2026 · by Polycarp

Every page on this site says the agent asks your team when it doesn't know. This is what that actually looks like, from both sides, because the trust in an AI agent lives in the moment it stops being one.

What the visitor sees

The visitor asks something your pages don't cover. The agent says so, in their language:

Our pages don't say, so I won't guess. I've asked the team, and their answer will appear here in this chat.

If they would like a person anyway, they can ask for one at any time, and they are asked for a name and an email so the answer can reach them if they have closed the tab. Nothing is invented in the meantime. The chat stays open; they can keep asking the things the pages do answer.

What your team receives

An email, to the address you set for that agent, with the whole conversation so far. If the visitor wrote in German, your team reads it in English. If you have connected Slack or your own system, the same conversation arrives there too, once, the moment it is handed over.

Inside the dashboard the conversation is in the inbox, marked as waiting, with the pages the agent used for the answers it did give, so your team can see what the visitor has already been told.

How the reply gets back

Someone on your team replies from the dashboard, under their own name. The reply appears in the visitor's chat, in the visitor's language. Once a person has taken over, the agent steps back from that conversation.

Nothing the team member types is charged as a conversation. A person's work is a person's work.

How the question stops being asked

This is the part that makes the agent better, and it is the only way it gets better. When your team answers a question the agent couldn't, Polycarp drafts that answer into your knowledge base: question, answer, source "your team". You approve it, edit it, or dismiss it. Once approved, the agent answers that question itself next time and cites it as coming from your team.

Nothing a visitor says ever becomes knowledge on its own. Only an answer your team gave and said yes to. So the agent's knowledge is a record of what your business decided, not of what strangers typed.

What this means for the trust question

An AI agent you can trust on your website is not one that never fails. It is one whose failure is visible, bounded and routed: the visitor is told, your team is told, the answer comes from a person, and the gap is closed by someone with the authority to close it. Try it on any sample business: ask the one thing its pages don't say.

All articles · Start your free month · Try a sample agent