Choose who answers: AI, humans, or AI when you're offline

Each bot has a Who answers setting on Chat → Bots. It decides whether the bot replies to a visitor, or whether the conversation waits in your Inbox for a person.

The four modes

• AI first, then a human (default) — the bot answers, and hands over when it isn't confident or the visitor asks for a person.
• AI only — the bot always answers and never escalates. Suitable for a pure self-serve help widget.
• Humans while we're open, AI otherwise — visitors reach your team while someone is online and you're inside your office hours. Outside that, the bot answers instead of leaving them waiting.
• Humans only — the bot never replies. Useful when you want a bot configured and ready but not live yet.

The bot always gets out of the way

Whichever mode you pick, one rule holds and cannot be switched off: the bot stops answering a conversation the moment a teammate steps in. Two things count as stepping in:

• replying to the conversation, or
• assigning it to yourself (or to anyone else).

From that point the bot adds nothing further to that thread. Unassigning does not restart it — resuming the bot is always a deliberate action, so a conversation can never quietly revert to a bot in the middle of a problem.

Take over, or hand it back

Open a chat conversation in the Inbox and you'll see a pill in the header:

• Bot handling — the bot is answering this visitor. Click Take over to silence it before you start typing, which is useful when you can see the conversation going wrong and don't want another bot reply landing while you compose.
• You have it — a person owns the conversation. Click Return to bot to let the bot answer again.

Both are recorded in the audit log, because handing a live customer back to a bot is a real decision the team should be able to see.

When a visitor asks for a person

Phrasings like "can I talk to a human", "connect me to support" or "I want a real person" route the conversation to your team and assign it, in every mode except AI only — where there is no human path by design, so the bot keeps helping rather than promising someone who never arrives.

This check is deliberately cautious: it looks for an actual request, not any mention of people. Asking "do you have a human resources module?" won't silence the bot. Anything it misses is still caught by the bot's own confidence handoff.

Changing a mode mid-conversation

A conversation keeps the mode it started under, so changing a bot's setting never changes the rules under a chat that's already underway. New conversations pick up the new mode immediately.