When does the AI answer, and when does it hand the conversation to me?

By default, your AI assistant answers new conversations on its own. It steps back and brings in a human in a few specific situations, and it always tells the customer when it does.

If a customer sends a few messages quickly, the AI answers the newest one, and its reply takes the earlier ones into account. If one of those earlier messages asked for a person, the conversation is still handed over.

How customers know they're talking to the AI

Every reply the AI writes in your website chat carries a small AI label, and a teammate's reply never does. SMS, WhatsApp, Messenger, Instagram and email have no room for a label, so there the AI's first reply in each conversation starts with "Hi, I'm the AI assistant for" and your business name. This is always on. On email, every AI reply also ends with: "This reply was written by our AI assistant. If you would like a person instead, just reply and say so."

When the AI hands over

• The customer asks for a person. Phrases like "talk to a human", "speak to an agent" or "transfer me" are recognized automatically, and the AI hands over right away rather than trying to talk them out of it. If they also ask whether they are talking to a real person or a bot, the AI answers that first, "Yes, I'm an AI assistant for" and your business name, and then hands over.
• The customer writes about danger or self-harm. A message about suicide or self-harm, a gas smell, a fire, or someone who cannot breathe gets a fixed safety message first: call or text 988 for suicide or self-harm, call 911 for any danger. Your own emergency message follows it if you have one, then the usual hand-off line, and your team is told at once. The "while you wait" offer below never follows a safety message. A burst pipe or a flood is urgent for you but not a 911 call, so it gets your emergency message, or the usual hand-off line if you have none.
• The AI itself decides a human is needed. For anything outside what it can confidently handle, it hands off with one fixed line: "Let me bring in a team member who can help with this. They'll reply here as soon as they can." It is the same line at every hour, so it never promises a time, and the AI marks the conversation as needing a human. On web chat with no pre-chat form, if we have no email or phone for that visitor yet, it adds one more sentence and opens the "how can we reach you?" box, so someone can still reach them if they leave: "If you might step away, leave your email below and we'll send the reply there too." when your inbox emails visitors their replies, or "If you might step away, leave your email below so the team can get back to you." when it does not. The safety message is always sent in French or Spanish when the customer writes in it. If you switched French or Spanish on, the hand-off lines, the "while you wait" offer and the "I don't have that" offer are too.
• The AI cannot answer a question. The first time, it says so and offers to bring in your team, and nobody is alerted yet. A "yes" to that offer, or a second question in a row it cannot answer, hands the conversation over as above. A question that asks for a person, or sounds urgent, is handed over straight away.
• A teammate takes the conversation over. Assigning it to a person by hand, or replying to the customer yourself, stops the AI answering that conversation. A conversation shared out to someone automatically keeps the AI answering. See How do I give a conversation to someone else?
• You pause or turn it off for that conversation. See How do I take over from the AI in the middle of a chat?

Email works a little differently

On an email inbox, the AI only emails your customer when that inbox is on Auto-pilot. On Suggestive or Off, a hand-off sends your customer nothing, not even the holding message when something breaks: the conversation simply comes to your team. The "while you wait" offer below is never sent by email.

How to spot a conversation that needs you

Conversations waiting on a human show a "Needs human" badge in your list, and you can filter your inbox to just these with the Needs human filter, see How do I filter my inbox, and save the filters I use most? The AI status panel inside an open conversation, opened from the info icon in the header, shows the reason it escalated and when.

What the customer sees while they wait

If nobody on your team has replied within about five minutes of a hand-off, the AI posts one more message on its own: "While you wait for the team, I can try to help. Just reply here and I'll answer, or keep waiting and a team member will be with you." It offers only once per conversation, and only after a hand-off it chose itself, never after a safety message and never after the "something went wrong" message below.

If the customer replies to that offer, the AI answers again using the whole conversation. The conversation still shows as needing a human and still sits in your queue exactly as before, so a teammate can step in at any point. If a teammate replies first, the offer never comes, or if it already has, the AI stays quiet.

If the AI genuinely breaks

On the rare occasion something goes wrong on our side, like a provider outage, rather than leaving the customer with no reply at all, the AI sends a short holding message saying a team member will follow up, and the conversation is handed to a human. This is different from a normal handover: it only happens when something has actually failed, not as part of a routine handoff, and it never offers to help again while the customer waits. On web chat with no pre-chat form it can add the same request for an email described above, if we still have no way to reach that visitor.