Keyword filtering
Silently drop visitor messages that contain words you don't want to see — spam, profanity, promotional URLs.
Keyword filtering catches spam and abusive content before it lands in your inbox or pings an agent. Configured from Chat → Settings → Blocked keywords.
How matching works
- One keyword per line in the settings textarea.
- Case-insensitive substring match against the message body —
casinomatchesCasino,BIG-CASINO, etc. - A match triggers a silent drop: the visitor sees a normal-looking success response, but nothing is persisted and no agent is notified.
What to put in there
- Spam keywords specific to your space ("crypto", "loan", "SEO services").
- Profanity if your widget runs on a family-friendly site.
- Competitor names if you keep getting "are you better than X?" from bots.
- Common phishing URLs if a campaign keeps targeting your domain.
Avoid generic words like "help" or "support" — those would silently nuke legitimate questions. Test a few terms by sending yourself a test chat from incognito.
Why "silent" drop
If the widget said "your message was filtered", a determined spammer would just retry with variations until they found a hole. By acknowledging silently, you don't reveal the filter exists or what it catches. The spammer thinks their message went through; no human ever sees it.
What's NOT filtered
- Agent messages — only visitor-side traffic is scanned.
- Internal notes — agents talking to each other are never filtered.
- File names — only message bodies are matched; attached file names aren't scanned.
For visitor IPs that need a stronger response, use the IP blocklist — that blocks the whole connection rather than individual messages.