Skip to main content
Every chat with your chatbot is a conversation. A conversation is in AI mode or in human mode. In AI mode, the chatbot answers. In human mode, the chatbot stays silent and a person on your team answers. This page explains how a conversation moves between the two modes, how your team closes it, and what the visitor sees at each step. To set up human support, see Set up human support. To reply to visitors, see Reply and take over.

​
AI mode and human mode

New conversations start in AI mode. You can change this in Settings > Chat Modes. If you choose Always Starts New Conversation with Human, every new conversation starts in human mode, and the AI does not reply to it. If you choose Follow Working Hours, new conversations start in human mode during your working hours. Outside them, they start in AI mode, or in human mode with an away message for the visitor. See Answer during working hours. If you choose Let the Visitor Choose, new conversations on the website widget and WhatsApp start by asking the visitor whether they want the AI or a person. See Let the visitor choose the AI or a person. In human mode, the visitor’s messages are saved and your team can read them in Chat History. The AI does not answer them. The AI starts to answer again only when the conversation goes back to AI mode.

​
How a conversation reaches your team

A conversation moves from AI mode to human mode in these ways. When the visitor, the AI, or the lead form starts the move, it is called an escalation.
  • The visitor chooses a person at the start. With Let the Visitor Choose, a new conversation starts with a question and two buttons. The person button escalates the conversation. In the website widget, the visitor confirms in a dialog first.
  • The visitor selects the human support button. When human support is on, a button can appear after each answer. You choose its label. The visitor confirms in a dialog before the conversation escalates.
  • The visitor asks for a person, and the AI decides to escalate. When human support is on, the AI follows your escalation policy. The default policy escalates when the visitor asks for a human, when the matter is urgent, or when the visitor is clearly frustrated after the AI could not help. You can replace this policy with your own rules.
  • The visitor submits the lead form. If Auto-escalate on lead capture is on in Leads > Leads Settings, a submitted lead form escalates the conversation.
  • Your team replies from the dashboard. When a team member replies in Chat History, the conversation moves to human mode so that the AI stops answering.
  • Your team takes over. A team member selects Take over the chat in Chat History, or Lock conversation in the … menu when your team holds the conversation without a lock, for example after a team reply or in a conversation that started in human mode. This moves the conversation to human mode and locks it there.
  • Your team replies from an AI assistant or API v2. If the AI still answers the conversation, or an escalated conversation waits for your team, the reply takes the conversation over first. This moves it to human mode and locks it there.
An escalation marks the conversation as escalated and sends a notification to your team, by email and to other places you set up. See Notifications. A reply or a take over from the dashboard does not send a notification, because a team member is already in the conversation.

​
Take over and release to AI

Take over is for a team member who wants to handle a conversation personally. After Take over the chat:
  • The conversation is in human mode.
  • The visitor cannot switch the conversation back to AI.
  • If the AI is writing an answer at that moment, it stops.
When you are done, select Release to AI. The conversation goes back to AI mode, and the AI answers the visitor’s next message. From the dashboard or the iPhone app, a reply without a take over moves the conversation to human mode but does not lock it. The visitor can still select Continue with AI to get AI answers while they wait for you. After that reply, the dashboard offers Release to AI, not Take over the chat. To lock a conversation that is in human mode without a lock, open the … menu at the top of the conversation and select Lock conversation. It does the same as Take over the chat: the visitor sees “A human agent has joined the conversation” and can no longer select Continue with AI. If an escalated conversation waits for your team to answer the visitor’s last message, the menu does not show Lock conversation. Select Take over the chat instead. You can also use POST /api/v2/chatbots/{chatbotId}/conversations/{threadId}/take-over. See Lock a conversation that your team holds. After a team member replies, SiteGPT can email the reply to the visitor. It does this if the visitor left an email address, did not come back to the chat after your reply, and the conversation is not resolved. The email has a link back to the conversation.

​
Resolve closes a conversation

When the matter is done, a team member selects Mark as Resolved. A resolved conversation is closed:
  • The visitor sees “This conversation has been closed.” and a Start a new conversation button. The visitor cannot send more messages in the closed conversation.
  • If the AI is writing an answer at that moment, it stops.
  • The conversation cannot be escalated while it is resolved.
Only your team can reopen a resolved conversation. Open it in Chat History and select Mark as Unresolved. SiteGPT does not resolve conversations on its own. A conversation stays open until someone on your team resolves it.

​
What the visitor sees

This table shows what the visitor sees in the website widget at each step.

​
How your team finds conversations

Chat History has four views:
  • Attention: conversations that are not resolved and that are escalated, unread, or have negative feedback from the visitor.
  • Open: all conversations that are not resolved. This is the default view.
  • Closed: resolved conversations.
  • All: every conversation.
A conversation can show the Escalated badge and the Resolved badge. To work through the views, see Review conversations.

​
Conversations on other channels

When you connect SiteGPT to a channel such as Slack or Facebook Messenger, the AI answers first on that channel. The handoff works differently from the website widget:
  • Your team usually replies inside that tool. Replies you write in the SiteGPT dashboard are not sent to these channels. WhatsApp is the exception: dashboard replies go to WhatsApp. See WhatsApp.
  • On channels such as Slack, Messenger, WhatsApp, and Zendesk Messaging, if you resolve a conversation in SiteGPT, the visitor’s next message starts a new conversation.
  • Each channel has its own way to hand a conversation to a person and back. For example, some channels hand the conversation back to the AI when your team closes it in that tool.
Zendesk Messaging works differently from the other channels. Each integration page describes how that channel behaves. See Integrations.