Scored, not matched
A single suspicious signal is usually nothing. Someone typing in capitals is not a scammer; someone typing in capitals, pinging everyone, and posting an image with a Nitro link probably is. So each message is scored against every pattern you have enabled, the scores add up, and the total decides what happens.
The patterns
Ten of them, each with its own weight:
- Higher weight — Nitro scams, Steam gift scams, prize and giveaway scams, crypto scams, and mass-ping spam.
- Medium weight — image scams, image-only posts, excessive repetition, and suspicious links.
- Lower weight — excessive capitals.
You pick which patterns are active. Attachment URLs can be checked as well as message text, which is worth leaving on — the image scam and the link in it are usually the same message.
Sensitivity
One setting moves every threshold at once. At low sensitivity a message needs several patterns to agree before anything happens; at high sensitivity a single high-weight match is enough to flag. The same setting governs the thresholds for logging, so turning it down makes the channel quieter as well as the enforcement gentler.
Exemptions
Roles you nominate are never checked, and individual member IDs can be exempted too. Worth knowing: the spam page’s safe-role list is a separate list from the moderators configured for impersonation detection — filling in one does not fill in the other.
Free and Pro
Spam detection and logging are free for every server. What Pro adds is acting on a detection — deleting the message and timing the author out — at $49/mo or $490/yr per server.