Before you start
Have an enabled assistant, a test contact in the connected GHL location, and a channel that can both receive and send messages. A channel toggle allows an assistant to use a connection; it does not create the provider connection itself.Create the routing rule
- Open Active Tags in the intended subaccount.
- Click New active tag or Create active tag in the empty state.
- Enter a Rule name that your team will recognize.
- Set the Matching tag value you will apply to contacts.
- Choose the assistant and save the rule.
- In GoHighLevel, add that matching tag to the test contact.
Configure reply behavior
Open the assistant’s Chat Settings:- Wait time delays the queued reply in seconds and helps group closely spaced incoming messages.
- Max responses limits AI responses in a conversation.
- Channels controls the channels the assistant may answer. Enable only the connected channels you intend to use.
- Sleep mode pauses eligible replies after a human sends a message. Sleep duration controls that pause. Sleep on workflow messages determines whether recognized automated sends also start it.
Chat Settings reference
Defaults describe a fresh form; saved assistants load their existing configuration.
Wait for the saved state before testing. Channel selection allows responses through an existing integration; it does not connect the channel or configure active-tag routing.
Verify a delivered reply
Send a new message from the test contact through the actual channel. Wait for the configured delay and inspect both sides of the conversation. Confirm the customer received the reply and that the intended assistant handled it. If there is no reply, check the matching contact tag,ai_off, assistant enabled state, channel settings, sleep state, response cap, integration connection, and funding/error status. Removing every matching active tag prevents this automatic routing path from selecting an assistant; it does not promise a particular memory or logging outcome.
For a manual pause, use ai_off. For an in-progress indicator, see AI Replying.