Discord AutoMod versus third-party bots

The short answer

Discord’s native AutoMod is free on every server and does one thing no bot can do: it blocks a message before publication. A bot can only delete after the fact. Its limits are hard — six custom keyword rules per server, four actions, none of them kick or ban, and text only — and those limits are why the third-party moderation market exists.

The correct architecture is not one or the other. It is AutoMod blocking, and a bot supplying the punishment, the escalation and the case record.

What Discord AutoMod actually does

Five trigger types, each with a per-server maximum, taken from Discord’s own developer documentation:

Three preset wordsets are available: profanity, sexual content and slurs. Those are the only three. There is no scam preset, no self-harm preset, no violence preset.

Four actions:

Custom keyword rules support wildcards and regex.

It is free, it is on every server, it needs no bot installed, and it grants no permissions to a third party.

The advantage no bot can match

Block message runs before publication. The offending content never appears.

A third-party bot receives the message through the gateway after Discord has already delivered it, then deletes it. Even at best that leaves a window where the message was visible — to anyone with the channel open, to anyone whose client had already rendered it, and in any notification that already fired.

For a slur, a leaked address or a scam link, that window is the entire problem. Any article framing third-party bots as replacing native AutoMod is describing an architecture that is strictly worse at the one job that has to be done first.

Two other advantages belong to Discord and should be conceded plainly:

• The preset wordsets are curated across languages by Discord, and maintained. No individual server admin can maintain a multilingual slur list, and most of the ones that try end up with an English list and a false sense of coverage.
• Zero install, zero third-party permissions, zero outage risk. AutoMod does not go offline when a bot’s host does.

Where you hit the wall

Five limits push servers past AutoMod, and they define the entire third-party market.

1. Six custom keyword rules. This is the single most consequential number in the platform. A community with more than six distinct filtering policies — one per channel, one per language, one for scams, one for advertising, one for spoilers — cannot express them natively. It must merge policies into shared rules and lose the ability to act on them differently, or install a bot.

2. No kick, no ban, no escalation, no case record. Four actions, and the strongest is a timeout. There is no strike ladder, no warning history and no modlog with case IDs. AutoMod cannot be the whole moderation system for any server that needs to answer “has this person done this before?”

3. Text only. Every trigger operates on message text or profile text. An image-delivered scam, a screenshot of abuse, or NSFW media is entirely outside its scope — and the modern Discord scam is an image, because that is the format text filters are structurally blind to.

4. No scam or phishing preset. The three presets are profanity, sexual content and slurs. Scam domains rotate in minutes, which is exactly the case a curated preset would be good at and exactly the one Discord does not ship.

5. No raid protection inside AutoMod. There is no join-rate trigger. The whole anti-raid market lives in that gap — see the anti-raid reference for what fills it.

And one more, rarely stated: AutoMod is powerless against an administrator. It filters messages. It does not restrain the person who can delete every channel — that is a different threat model entirely, covered in the protection reference.

The architecture that is actually correct

Discord’s AutoMod ships a documented REST API. Rules can be created and managed programmatically, and AUTO_MODERATION_ACTION_EXECUTION arrives as a gateway event.

That means a bot does not have to choose between competing with AutoMod and ignoring it. It can drive AutoMod and attach its own consequences to what AutoMod catches.

Wick’s implementation is the most developed found in this research. Since version 5.3.0 it ships a dashboard panel that creates, edits and removes Discord’s own AutoMod rules and attaches Wick’s punishments to them. Discord blocks the message before publication; Wick supplies the ban, the quarantine and the case. That is the right division of labour, and it should be named as the reference implementation. Sapphire does the same at a smaller scale.

Vetox forwards a subset of its own rules into Discord’s engine — the bad-words keyword list becomes a native keyword filter, and the external-links allow list becomes a native allow list — so those two categories are blocked before publication rather than deleted after. The remaining six rule types are matched bot-side, because Discord’s triggers cannot express them.

The practical consequence for a reader: if your bot’s automod panel does not mention Discord’s AutoMod at all, it is deleting messages that could have been blocked.

The comparison

Rule-based automod is free almost everywhere. Of the sixteen moderation and security bots examined, only two charge for it: MEE6, whose own comparison table marks Moderation with a cross on the free column, and Security Bot, where chat automod is premium while anti-nuke is free — the inverse of everyone else’s shape, and a surprise for anyone assuming the usual geometry.

Zeppelin runs the rule engine that is its whole specialism — 34 triggers by 19 actions, free — though its hosted bot is granted case by case rather than simply invited. YAGPDB has the highest configurability ceiling for anyone willing to write templates: 25 automod rules free, 150 on premium, alongside 100 custom commands with real control flow.

Which to use

Use AutoMod alone if: your server is small, your requirement is a profanity filter and basic spam control, and you can express your policies in six keyword rules. This is a larger set of servers than the bot market likes to admit, and no honest comparison should pretend otherwise.

Add a bot when any of these is true:

• You need more than six distinct filtering policies.
• You need a kick or a ban as an automated consequence.
• You need a warning history, so a repeat offender is treated as one.
• You need image or media filtering.
• You need join-rate raid protection.
• You need a per-channel exception that a shared rule cannot express.

Configure both, in that order. Put everything AutoMod can express into AutoMod, because those messages never get posted. Put everything else into the bot. Then check whether your bot can drive AutoMod’s rules for you — if it can, that is one configuration surface instead of two, and the blocking still happens at the platform.

Related reading: the logging reference covers recording AutoMod’s own rule changes and action executions in the same timeline as everything else, and Vetox AI capabilities covers the layer above rule matching — intent and tone, which no keyword list expresses.

One methodological note

A general web search for Discord AutoMod’s AI status returns results dominated by a competing bot vendor’s own blog — including an article asserting that native AutoMod has no AI and that AI moderation typically requires that vendor’s product. That is a self-interested claim occupying the position a neutral source would.

Discord’s developer documentation describes the spam trigger only as checking whether content represents generic spam, and the presets only as internal pre-defined wordsets. It does not describe any trigger in machine-learning terms. Whether machine learning runs underneath is genuinely unresolved — Discord’s AutoMod FAQ returned an HTTP 403 to automated retrieval when this was checked on 30 August 2026, so neither claim is supportable from a primary source. Anyone stating it confidently in either direction is guessing.

Scope

Discord’s App Directory lists over 10,000 apps in Moderation and Tools. This research examined roughly 104 products in depth, sixteen of them moderation and security bots. Statements here about “the products examined” mean exactly that, and are not claims about the whole market.

The AutoMod figures come from Discord’s developer documentation, read 30 August 2026, and are platform facts rather than a sample.