Let visitors talk to your team

​

Before you start

• Plan: any plan. See Plans and limits.
• Role: Manager or higher. See Team members.
• Decide who on your team answers escalated conversations, and when.

​

Turn on human support

​

Set the escalation buttons

​

Tell the AI when to escalate

1. Under Escalation Policy, write your rules in When should the AI escalate? Use plain language. The limit is 2,000 characters.
2. Select Save Changes.

Escalate only when the visitor explicitly asks to speak with a human, or asks about pricing, demos, quotes, or appointments. In all other cases, answer the question.

​

Choose who answers new conversations

​

Answer during working hours

• SiteGPT decides how a conversation starts when the visitor sends the first message, at that time in your time zone.
• The hours decide the mode of new conversations only. A conversation keeps its mode when the hours start or end.
• With Your team answers later, SiteGPT sends the away message when a visitor writes outside working hours in any conversation in human mode. This includes conversations that started during the hours, escalated conversations, and conversations where a team member replied. Each conversation gets the away message one time. A conversation that a team member took over gets no away message, and neither does a closed conversation.
• Working hours apply to the website widget, the API, and the chat integrations. Zendesk and Zoho SalesIQ do not use working hours. In these two tools, new conversations start with the AI or with your team as set by the last fixed option you saved, which is the AI by default. While Let the Visitor Choose is on, they start with the AI.
• If you later select Always Starts New Conversation with AI, Always Starts New Conversation with Human, or Let the Visitor Choose and save, working hours turn off. SiteGPT keeps the schedule, and it shows again when you select Follow Working Hours. Setting only the chat mode to AI or human mode from the CLI, the API, or the MCP server also turns working hours off. In the MCP server, update_sitegpt_settings with chatMode set to WORKING_HOURS turns the saved schedule on again. See Appearance and settings tools.

​

Set working hours with API v2

{
  "workingHours": {
    "enabled": true,
    "timezone": "America/New_York",
    "days": {
      "mon": [{ "start": "09:00", "end": "17:00" }],
      "tue": [{ "start": "09:00", "end": "17:00" }]
    },
    "outsideHours": "AWAY",
    "awayMessage": "We are away right now. We will answer by email."
  }
}

• timezone is an IANA time zone name. Times are HH:MM. A day that you leave out is closed.
• outsideHours is AI (the AI answers) or AWAY (your team answers later). AWAY needs an awayMessage.
• "workingHours": null turns working hours off and deletes the schedule. Sending only mode turns working hours off and keeps the schedule.
• A workingHours object with enabled: true turns visitor choice off.
• enabled: false saves the schedule and keeps working hours off.
• If you send mode together with workingHours, the schedule stays as you sent it.
• A GET of the section also returns currentMode, the mode that a new conversation gets right now. You cannot set it.
• A GET also returns mode, the fixed chat mode. Zendesk and Zoho SalesIQ use mode, and SiteGPT uses it when working hours and visitor choice are off.

​

Let the visitor choose the AI or a person

• Website widget and chat link: A new conversation shows your question below the welcome message, and the two buttons where your conversation starters usually show. The AI button hides the question for this conversation, and the visitor can type. The visitor can also type a message without a button, and the AI answers. The person button opens the same confirmation dialog as the escalation button. After the visitor confirms, the conversation goes to your team. If you set a Human Support Confirmation Message, the chat shows it. If you did not, the chat shows only that the conversation changed to human mode, so set one. See What the visitor sees. Each new conversation asks again.
• WhatsApp: The visitor writes first. SiteGPT does not answer the first message. It sends your question with the two buttons. See Connect WhatsApp.
• Other channels: Messenger, Slack, Google Chat, Crisp, Freshdesk Messaging, Zoho SalesIQ, Zendesk, and the API do not ask the question. New conversations there start with the AI, also when the last fixed option you saved was Always Starts New Conversation with Human. In API v2, a new conversation that you create with a mode starts in that mode.

​

Set visitor choice with API v2

{
  "visitorChoice": {
    "enabled": true,
    "question": "Would you like to chat with our assistant or talk to our team? We reply Monday to Friday, 9 to 5.",
    "aiLabel": "Chat with AI",
    "personLabel": "Talk to our team"
  }
}

• The first write must include enabled. Without it, the request fails with “Visitor choice needs enabled set to true or false.”
• Only the fields that you send change. The other fields keep their saved values. { "enabled": true } turns visitor choice on with the saved text, or with the default text if none is saved.
• An empty text or null in question, aiLabel, or personLabel sets the field back to the default. A GET returns null for a field that uses the default.
• "visitorChoice": null turns visitor choice off and deletes the text. enabled: false turns it off and keeps the text.
• enabled: true turns working hours off, unless the same request also sends workingHours: then what you send for workingHours is kept. Working hours that stay on take priority, and visitor choice does not show. So when you send the whole settings object back, set workingHours.enabled to false yourself. The same applies the other way: turning working hours on turns visitor choice off only when the request does not also send visitorChoice.
• Sending mode turns visitor choice and working hours off, unless the same request also sends them.
• If the two labels would be the same, the request fails with 400 VALIDATION_FAILED. A label over 20 characters or a question over 1,000 characters also fails.
• While visitor choice is on, currentMode is AI. The saved mode does not change.

​

Other ways a conversation escalates

• After a lead form. Turn on Auto-escalate on lead capture in Leads > Leads Settings. See Collect leads from conversations.
• From the visitor choice. With Let the Visitor Choose, the person button at the start of a conversation escalates it. See Let the visitor choose the AI or a person.
• From a follow-up suggestion. A follow-up suggestion can be an escalation button. See Escalate buttons.

​

What the visitor sees

1. The visitor selects the escalation button, or the AI decides to escalate.
2. If the visitor selected the button, the widget asks them to confirm.
3. The chat shows that the conversation changed to human mode. Then it shows your confirmation message.
4. The chat asks for contact details if lead collection is on, or if Pre-Chat User Details is Required or Optional. Your team can then follow up by email. See Collect leads from conversations.
5. The AI stops answering. The visitor’s messages wait for your team.

​

Get notified when a conversation escalates

​

Check that it works

1. Open your chatbot on your website or in the dashboard preview.
2. Ask a question. After the answer, select the button with your Request Human Support Prompt text, then confirm.
3. The chat shows your confirmation message.
4. In the dashboard, open Chat History. The conversation shows the Escalated badge and is in the Attention view.