Message Guard: filter spam and unwanted messages
Stop irrelevant, personal, spam, abusive or sensitive messages before they reach your AI assistant, and review what was blocked.
Message Guard filters irrelevant, personal, spam, abusive and sensitive messages before they reach your AI assistant. Blocked messages are still shown in the conversation and do not use your conversation quota.
Open it
- Web: Settings > Message Guard, with two tabs: Settings and Activity.
- Mobile: Home > Settings > Message Guard.
Settings
| Setting | What it does |
|---|---|
| Enable Message Guard | On by default. Turn off to send every message straight to the AI. |
| Use AI classifier | When fast rules cannot decide, a lightweight AI model classifies the message. When off, only rules apply. |
| Always block these categories | Force-block a category even if the classifier would allow it: Personal, Story reaction, Political, Abusive, Inappropriate, Spam, Irrelevant, Sensitive. |
| Allow keywords | Messages containing any of these always pass (comma-separated, for example booking, invoice, warranty). |
| Block keywords | Messages containing any of these are always blocked (for example loan offer, seo service). |
| Mute non-customer conversations | Off by default. When the guard is confident a contact is a personal acquaintance and not a customer, it turns the AI off for that conversation and stops saving their messages. It can be reversed by turning the assistant on again. |
Choose Save.
Activity
Guard activity lists every message the guard blocked, ignored or held for review, with the action (Review, Blocked, Ignored, Allowed) and whether a Static rule or the AI classifier decided. Filter by Needs review, Blocked, Ignored or All, and choose Open conversation to see the context.
How the guard understands your business
The guard learns your business context automatically from your assistant settings (business name, instructions, hours, FAQ topics), so you do not type it.
Tips
- Add the words your real customers use (for example "price", "appointment") to Allow keywords.
- Check Activity during the first week to make sure no genuine customer is blocked.
Troubleshooting
- A real customer was blocked: open it in Activity, add a relevant word to Allow keywords, and reply manually from the Inbox.
- Too much spam still reaches the AI: add Block keywords or turn on Always block for the relevant categories.
Limitations
- Blocking a message is different from blocking a contact. To ignore a specific person completely, use the channel's Blocked Users or Blocked Numbers.
- Sticker messages and emoji-only story reactions are dropped silently before reaching the guard.
Last updated on
AI-to-human handoff
Let customers reach a person: set up contacts and triggers, receive human support requests, take over, and return the chat to the AI.
Notifications: push, in-app and email
Choose how you are alerted about new leads and human support requests, enable push notifications, and review alerts.