The Auto-moderation module
Auto-moderation does not replace moderators: it takes away the work that never needed a human. Set well, it is invisible. Set badly, it hits normal conversation and teaches members to work around it.
Updated 12 September 2026
What Discord already does, and what it does not
Discord ships a free native AutoMod (Server Settings → AutoMod) that blocks keywords, invite links to other servers and mention spam, before the message is even posted. Turn it on: many servers never have.
What it does not do is respond over time. It deletes the message and stops there. It does not count repeats, does not escalate, and keeps no usable record.
That is where the Kuratos Auto-moderation module sits: it picks up afterwards rather than redoing the same thing.
The principle of graduated sanctions
Deleting a message changes nothing about the behaviour of whoever sent it. What changes it is that the third offence costs more than the first.
A sensible ladder: silent deletion, then warning, then short mute, then long mute. It handles the overwhelming majority of cases on its own, and the rare ones that go all the way genuinely deserve a human.
Set the counting window. Three offences in five minutes have nothing to do with three offences in six months — without a window, a long-standing member ends up sanctioned for a total spread over a year.
Word filters: the too-broad trap
A filter that is too broad is worse than no filter. It hits words inside other words, cuts legitimate conversation, and above all teaches your members to misspell deliberately — after which the filter is useless and you can no longer see anything.
Filter little, and only what is genuinely unacceptable. The rest is human judgement, precisely because it depends on context.
- Prefer whole phrases to isolated words.
- Exclude channels where the subject justifies blunt vocabulary, if you have any.
- Exclude staff roles: a moderator quoting a reported message must not be sanctioned for quoting it.
- Review the list every few months — a filter added after one specific incident often has no reason to survive.
Links and spam: where the gain is clearest
This is the part that pays most for the least effort. Scams almost always travel through a URL shortener, because the real address would be recognised. Blocking them takes a minute and removes most of the problem.
For spam, two caps are enough: mentions per message, and the same message repeated across channels — the signature of the account touring your server.
How to check your settings are right
- Start with the delete-only action and no escalation: you see what the rule catches without sanctioning anyone.
- Watch the log for a week. If you see normal messages in it, the filter is too broad.
- Check staff roles are excluded, otherwise moderation sanctions itself.
- Test with a second account: a filter that never fires produces no error, it simply does nothing.
Key points
- Turn on Discord’s native AutoMod first: it is free and acts before posting.
- The module’s contribution is graduated sanctions and the record, not the deletion.
- A too-broad filter hits normal conversation and teaches evasion.
- Blocking URL shorteners removes most scams, in one minute.
- Exclude staff roles, or moderation sanctions itself.
Frequently asked questions
Does auto-moderation replace moderators?
No. It handles what needs no judgement — spam, links, mechanical repeat offences — to leave time for conflicts between people, which do need it.
Should you use Discord’s AutoMod or the Kuratos module?
Both. AutoMod acts before the message is posted, which no bot can do. The Kuratos module picks up afterwards: repeat counting, graduated sanctions and a readable log.
How do you stop the filter hitting normal messages?
Filter whole phrases rather than isolated words, exclude staff roles and the relevant channels, and read the log during the first week: that is when false positives show up.
Put this guide into practice
Kuratos sets up in minutes what this guide describes, and is configured from a web interface rather than through commands.