# How to set up auto-moderation on Discord with AI

> Have AI read each message and delete spam, warn members or time them out for abuse. Set up AI automod for your Discord server from a template, no code.

Source: https://buildcord.dev/guides/discord-ai-automod/
Updated: 2026-11-05
Publisher: Buildcord (https://buildcord.dev)

Word filters catch the words you list. An AI check reads what a message means, so it catches the scam that's spelled oddly and lets a friendly joke through. The automod template sends each new message to the **Decide with AI** step, which sorts it into options you name, such as Fine, Spam or Toxic, and the bot deletes, warns or times out depending on the answer. It reacts to messages, so it runs on Buildcord Boost, the $8-a-month plan.

## Steps

1. **Turn on the Message Content intent.** In the Discord Developer Portal, open your application's **Bot** page and turn on **Message Content Intent**. Without it, Discord doesn't let a bot read what members write.
2. **Start from the automod template.** Open the automod template in Buildcord. It makes a bot with a **Message sent** flow, the AI check and what to do for each answer already drawn.
3. **Say what counts as a problem.** Open the **Decide with AI** step. Each option has a name and what it means, such as `Spam: ads, scams or the same thing over and over`. Write them in your server's terms. Each option gets its own way out of the step.
4. **Choose what happens.** Connect each way out to steps: **Delete a message**, **Direct message** the member with the reason, or **Timeout a member**. Leave Fine connected to nothing, so ordinary messages pass.
5. **Skip people it shouldn't judge.** Put a **Role** or **Permissions** check at the start so moderators are left alone, and a **Member check** to skip bots.
6. **Deploy and test.** Paste your token, deploy and send a test message from a second account. The bot's logs show each run, the option the AI picked and how sure it was.

## Good to know

- The step reads up to 8,000 characters. Afterwards `{ai.choice}` is the option picked and `{ai.confidence}` how sure it was, from 0 to 1.
- Act only when it's sure: add a **Comparison** on `{ai.confidence}` and send unsure cases to a moderators' channel instead.
- Choose the **Careful** model for closer reading, or **Fast** for busy servers.
- Cheap checks can come first. A **Message check** catches links, server invites, listed words, mostly capitals or a pattern without asking the AI.
- To keep a record of what happened, pair it with the [warning system](https://buildcord.dev/guides/discord-warning-system/) or a [server log](https://buildcord.dev/guides/discord-log-deleted-messages/).

## Questions

### Does it replace Discord's built-in AutoMod?

It works alongside it. Discord's AutoMod blocks keywords before a message is posted. The AI check reads messages after they're posted and judges what they mean, so the two catch different things.

### Will the AI make mistakes?

Sometimes. Write clear option descriptions, act only above a confidence you choose, and send borderline messages to moderators rather than deleting them. Every decision is in the bot's logs.

### Why does AI automod need a paid plan?

To see messages as they're sent, a bot has to stay connected to Discord all the time, which needs a machine of its own. That's what Buildcord Boost pays for.

## Build it

Start from the [Auto-mod template](https://buildcord.dev/templates/automod/), with every flow already drawn. Use the automod template: https://buildcord.dev/templates/automod/