Setting Up AutoMod: Six Filters and the Escalation Behind Them

Automatic moderation has a basic problem: it is either too loose and catches nothing, or too strict and punishes people for nothing. Both cost you members, just in different ways.

The way out is not to fine-tune detection, it is to choose the response separately per filter.

The six filters

Each filter runs independently and has its own escalation step.

Link blacklist. Blocks specific domains. The classic defense against advertising for other servers and against known scam sites.

Word filter. Blocks specific words. One detail that did not work for a long time and has since been fixed: words with punctuation before or after them were never matched. A word wrapped in spoiler bars slipped through silently, with no error visible anywhere.

Mass mentions. Catches messages that ping many users or roles at once.

Caps spam. Messages written mostly in capital letters.

Message flooding. Too many messages in too short a time from the same person.

Scam patterns. Detects typical scam setups, for instance the crypto giveaway DM solicitation along the lines of "I'm giving out $400 worth of solana, message me to get it". On top of that, unsolicited server advertising, meaning an invite link combined with a community reference, which is caught even when the wording was altered to dodge the word filter.

The escalation is the real lever

Per filter you can choose: delete, delete plus an automatic warning, timeout, kick or ban.

The mistake almost everyone makes at the start is setting the same penalty everywhere. But this is precisely where harmless and real violations can be separated.

A workable starting split:

Filter	Response	Reasoning
Caps spam	delete only	Almost always excitement, not intent
Message flooding	delete plus timeout	Actively disruptive, stops after a pause
Mass mentions	delete plus warning	Rarely accidental
Word filter	delete plus warning	Depends heavily on your word list
Link blacklist	delete plus warning	Can be an oversight
Scam patterns	ban	Nobody posts this by accident

The last row matters most. With the scam filter there is no plausible reason for a false positive on a genuine member, and the account behind it is usually compromised anyway.

Set exemptions before they hurt

Roles, channels and categories can be exempted. Three exemptions practically every server needs:

• Your staff from most filters, otherwise a moderator quoting a violation gets punished for it
• The advertising channel, if you have one, from the link blacklist
• Bot channels from message flooding, where rapid output is normal

The notice to the member is required, not decoration

When a message disappears and nobody says why, it reads as a bug. So AutoMod always shows the affected member a short explanation in the channel, and optionally the same explanation as a direct message.

The channel version disappears again after a few seconds so the notice does not become spam itself. The DM stays. Both can be designed in the Message Builder, with the reason and the server name in a locked block: a redesigned message that no longer says what actually happened would be worthless to the recipient.

The log channel is your control

AutoMod has its own log channel, and in the first weeks it matters more than any setting. It shows you what is actually being caught, and only that tells you whether your thresholds are right.

The honest order is: start loose, read the log, then tighten. Starting the other way round loses you members to false positives before you notice the setting was too strict.

Every setting in detail is in the AutoMod documentation.