Approve what the agent proposes审批助手提出的建议
The assistant can suggest creating a parent, a child or a label — but a person confirms it before anything exists.助手可以建议新增家长、孩子或标签——但要有人确认，记录才会真的存在。
这篇指南的正文目前只有英文版。标题、摘要与导览已翻译；内文仍在翻译中。
The assistant never writes to your records on its own. When a conversation gives it enough to create something, it queues a proposal and a human decides.
Where they appear
In Marketing → Inbox, inside the contact's own drawer.
Open a conversation and look at the profile button in the top right. A red dot on it means the assistant is waiting on you. Click it, and the proposals sit in the drawer alongside everything else known about that family — which is the point: you are approving a new record with the family's existing details in front of you.
What it can propose
| Proposal | What confirming does |
|---|---|
| Label a contact | Adds a short tag, so the queue can be filtered by intent |
| Remove a label | Takes a tag off once it has gone stale. Labels describe what someone wants *now*, so clearing one is as useful as adding it |
| Create a parent | Creates the parent record from what they gave in the chat |
| Create a student | Creates the child, attached to that parent |
| Book a class | Books the session — a make-up, a schedule change, or a trial slot |
| Request a package extension | Creates the extension request, carrying the photos the family already sent in the chat |
| Schedule a follow-up | Queues a message to send on a named day, when the family asked to be contacted later |
Approving
Read the proposal against the conversation above it, then approve or decline.
Approving creates the record under your name, not the assistant's. It is your decision on the audit trail, which is the point — the agent suggested, a person agreed.
Declining
Decline anything wrong or premature. Nothing is created, the conversation carries on, and the assistant does not re-propose the same thing at you.
Why it works this way
An assistant that could create records unsupervised would eventually create a wrong one at 2am, and nobody would know which of yesterday's fifty records to distrust. Queuing them keeps the speed — the reply still went out instantly — while a person stays responsible for what ends up in your data.
Related: Set up the AI agent · Work the WhatsApp inbox · Schedule follow-ups.