How AI Moderation Works on Discord (and What Noxyr Does Instead)

The three ways moderation can work

What an AI moderation tool does step by step

1. Receive the message. The bot or platform reads new text the moment it is posted.
2. Score it. A model returns a probability for categories such as spam, scams or abuse.
3. Compare to a threshold. You (or the vendor) decide how sure it must be.
4. Act. Delete, warn, time out or flag for a human.
5. Review. People check the flags and correct mistakes.

The hard part is step 3. Set it low and normal members get hit. Set it high and spam gets through. Context makes it worse: a joke between friends and a threat can look alike to a model.

Discord's own AutoMod

Discord ships a moderation tool in every server, found under Server Settings, then AutoMod. According to Discord's support article, it offers keyword filters, a mention spam limit and a spam content filter. A rule can block the message, send an alert to a channel for staff, or time the member out, though the options differ by filter. Rules are edited on desktop.

It is a good first layer because it runs inside Discord with no bot to add. Its limits: the keyword filters only know the words you give them, and Discord notes that the spam content filter does not catch every kind of spam, such as the same text repeated across messages and channels.

What Noxyr does: rules, not AI

Noxyr has no AI moderation feature. It does not read your members' messages with a model to judge them. Its moderation is rule-based:

• Anti-spam limits floods of messages, mentions and links.
• Anti-link blocks invite links and unwanted domains.
• Anti-nuke stops mass bans and channel deletions by a compromised account.
• Warnings give staff a clear escalation path.

The reason is simple. Rules do what you tell them and can be explained to a member who asks why they were punished. For many servers that matters more than catching every paraphrase.

Noxyr does have AI in other places: AI chat replies, ticket help and a builder. Those answer questions and set things up. They do not decide who gets moderated. There is also /ai check, a word-pattern heuristic that gives staff a hint about whether text was AI-written; it is a hint, not proof, and it is not moderation. See AI message check.

A practical stack

1. Turn on Discord AutoMod for your banned words and mention limit.
2. Add Noxyr anti-spam and anti-link for floods and unwanted links.
3. Add anti-nuke so a hijacked staff account cannot wreck the server.
4. Use warnings so humans handle repeat cases.
5. Keep people in the loop for everything borderline.

Questions to ask before trusting any AI moderator

• Where does the message text go? A model usually runs on someone else's servers. Check the privacy terms.
• Can you see why it acted? If a member appeals, you need a reason you can read.
• What is the default action? Flagging for review is safer than deleting or banning automatically.
• How does it handle other languages and slang? Many tools are strongest in English. Discord notes some of its own lists are English only.
• Who is accountable? You are. A tool cannot take responsibility for a wrong ban.

Why rules still win for many servers

Most server damage comes from a few patterns: floods, scam links, mass mentions and staff account takeovers. Those are countable. A rule that says "more than a set number of messages in a short time" needs no model and does not argue about context. Where judgment is needed, such as a heated argument, a human moderator with a warning system gives better results than a score.

More on moderation

For the full spam setup read the anti-spam guide, and for the warning system the warn system guide. If an account is already compromised, see what to do.