How to Tune Discord AutoMod So It Stops Flagging Normal Messages

To fix Discord AutoMod false positives from banned words, open Server Settings → AutoMod, audit your custom keyword filters for over-broad terms, narrow or whitelist the words that match innocent text (like "ass" matching "assist" or "class"), and tune spam and mention rules instead of relying on a wide blocklist. Most false positives come from short root words and * wildcards that match parts of longer, harmless words.

If your members keep getting messages deleted for saying things like "grass," "Scunthorpe," or "class assignment," your AutoMod isn't broken — it's matching substrings inside normal words. This is one of the most common AutoMod complaints, and it's fixable once you know where to look. Here's how to do it without leaving your server unprotected.

Why AutoMod flags innocent messages

Discord AutoMod works by matching text against rules you define: keyword lists, preset spam categories, mention caps, and link filters. It does not understand meaning. When a rule says "block the word ass," AutoMod will happily delete "password reset," "class starts at 8," and "I'll assist you." This is the classic "Scunthorpe problem" — a clean word containing a dirty substring.

The three biggest sources of false positives are:

• Short root words in your custom keyword list (3-4 letters) that appear inside many longer words.
• Wildcards (the * character) that intentionally widen a match and almost always widen it too far.
• Preset filters turned up to maximum sensitivity, which flag normal slang, sarcasm, or heated-but-fine debate.

Every one of these is a setting you control. If you're still setting AutoMod up from scratch, start with our guide on how to set up Discord AutoMod, then come back here to tune it.

How wildcards and presets over-match real words

A wildcard tells AutoMod "match this fragment anywhere." If you add scam*, you'll catch "scammer" — but also "scampi" and "scamper." Add *hole and you'll moderate "wormhole," "loophole," and "manhole." Wildcards feel powerful because they catch variations, but they're a leading cause of over-blocking.

Discord gives you three matching positions for keywords:

• word (no wildcard) — matches the whole word only, surrounded by spaces or punctuation. Safest.
• word* — matches anything that starts with "word."
• *word* — matches "word" anywhere, even inside another word. Most dangerous.

The preset filters (Discord's built-in lists for things like sexual content, slurs, and severe profanity) are maintained by Discord and update automatically, so they're generally safer than your own keyword lists. The catch is that they're broad by design. If a preset is catching ordinary conversation, the fix is usually to scope it to fewer channels or pair it with an allow-list, not to delete it entirely.

Step 1: Audit your current keyword filters

Open Server Settings → AutoMod → Block Custom Words (or whichever rule you created). Read your keyword list and ask one question per entry: could this fragment appear inside a normal word?

Flag every term that is:

• Three or four letters long with no wildcard boundaries.
• Using *word* (matches anywhere).
• A common syllable like "cum," "ass," "hell," "dam," "sex," "rape" — all of which live inside dozens of clean words (document, assist, hello, Amsterdam, Essex, grape, and so on).

Make a list of suspects. Don't delete anything yet — you'll decide in the next steps whether to remove it, narrow it, or whitelist around it. For a full reference on which settings belong in a healthy config, see our breakdown of the best Discord AutoMod settings.

Step 2: Narrow over-broad keywords

For each suspect term, pick the least aggressive match that still catches the real offense:

• Replace *ass* with the exact words you actually care about (asshole, jackass) as whole words. Now "assist" and "class" are safe.
• Replace scam* with scammer, scamming, and scam as discrete entries if you want precision, or keep scam* only if your community never says "scampi."
• Remove three-letter roots entirely unless they are slurs that have no clean homograph.

The rule of thumb: whole-word matches by default, wildcards only when you've consciously accepted what else they'll catch. It's more entries to type, but it's the difference between a filter members forget exists and one they complain about daily.

Step 3: Whitelist exceptions safely

Discord AutoMod lets you add an allow-list (also called "exempt words") to any keyword rule. This is the surgical fix when you must keep a broad term but a few clean words keep tripping it.

Say your community is about firearms and you block shoot* to catch threats, but "shooting range" and "photoshoot" keep getting deleted. Add shooting range and photoshoot to that rule's allow-list. AutoMod will skip the message when the exempt phrase is present.

Two cautions:

• Allow-lists are per-rule, not global. If three rules could catch the same word, you may need to add the exception to each.
• Don't allow-list so much that you hollow out the rule. If you're adding twenty exceptions to one keyword, that keyword is the wrong tool — narrow it instead (Step 2).

You can also exempt roles and channels. A staff-only or NSFW-flagged channel often shouldn't run the same word filter as #general. Use the rule's "Exempt Roles" and "Exempt Channels" fields to carve those out.

Step 4: Tune spam and mention limits without over-blocking

False positives aren't only about words. AutoMod's Mention Spam and Spam Messages rules can flag legitimately active users.

• Mention limit: Discord lets you cap mentions per message. Setting it to 3 will flag a mod who pings four people in a coordination post. For most communities, 5-6 is a sensible ceiling that still stops mention-raids. Reserve a low cap (2-3) for large public servers where mass-ping abuse is common.
• Spam / repeated text: The built-in spam preset catches copy-paste flooding, but enthusiastic fans posting "GG GG GG" after a match can trip it. If that's hurting your vibe, lean on a slowmode in hype channels instead of a hard AutoMod block, so messages slow down rather than vanish.
• Block action: Decide whether a rule should delete + alert, delete silently, or timeout the user. Auto-timeouts on a noisy keyword rule are how you accidentally mute regulars. Start with "alert a log channel only," watch what it would have caught, then escalate.

Step 5: Test changes before going live

Never trust a filter you haven't tried to break. Before you consider tuning finished:

1. Create a private #automod-test channel and exempt your test role from nothing — you want to see real behavior.
2. Type the edge cases you're worried about: "class," "assist," "grass," "Scunthorpe," your members' usual slang, and the actual bad word you're targeting.
3. Confirm the bad word still gets blocked and the clean words pass. If a clean word still dies, revisit the wildcard or add it to the allow-list.
4. Route AutoMod alerts to a log channel so you can review what's being flagged over the next few days. Patterns in that log tell you exactly which rule to loosen.

This short test loop catches most of the embarrassing false positives before your members ever hit them.

Step 6: Switch to AI moderation when keyword matching isn't enough

Keyword AutoMod has a hard ceiling: it cannot tell the difference between "I'll kill you at chess later" and a real threat, because it doesn't read intent. You can narrow and whitelist forever and still be playing whack-a-mole, because language is contextual and AutoMod is literal.

This is where context-aware AI moderation changes the math. Instead of matching a fixed blocklist, PeakBot reads what a message actually means and adapts per channel — so "this boss is going to murder me" in #gaming reads as banter, while a genuine threat in #general gets actioned. It understands sarcasm, gaming slang, and the difference between an insult aimed at a game and one aimed at a person, which is exactly the gap that produces AutoMod false positives.

PeakBot is a free, AI-powered Discord bot, and AI moderation is one of its 30+ features that are free with no time limit and no trial period. You can run it alongside Discord's native AutoMod (keep AutoMod for hard scam-link blocking, let the AI handle the judgment calls) or let it take over the keyword headache entirely. For a deeper look at the tradeoffs, read our AI Discord moderation guide.

To be fair to the alternatives: Discord's native AutoMod is genuinely good for what it is — fast, free, server-side, and reliable at blocking known scam links and Discord's maintained slur presets. MEE6 (premium $11.95/mo) and Dyno (premium $4.99/mo) both have mature rule engines if you prefer classic keyword moderation with a polished dashboard. The reason to add AI isn't that those tools are bad — it's that no keyword engine, however well-tuned, can read intent.

Step 7: Keep tuning as your community grows

Moderation is never "set and forget." A 50-member server and a 5,000-member server need different settings, and the same server needs different settings as its culture shifts.

• Review your AutoMod log monthly. New slang, new inside jokes, and new abuse patterns all show up there first.
• Re-check mention and spam caps after growth spurts. What felt safe at 200 members may throttle a thriving 2,000-member chat.
• Prune dead keywords. If a term hasn't matched anything real in months, it's probably only catching false positives.
• Watch the log as new members arrive. The edge cases you didn't think of will turn up on their own — your job is to watch and adjust.

PeakBot is already powering 500+ Discord communities and can replace MEE6, Carl-bot, Dyno, and TidyCord with one bot, so the ongoing tuning lives in a single dashboard instead of four. If you're rebuilding moderation from the ground up, you can also describe your ideal setup in plain English and let the AI Server Builder generate the channels, roles, and automations for you — though that specific feature is part of Pro ($8.25/month, or $69/year).

FAQ

Why does Discord AutoMod block words like "class" or "grass"?

Because those words contain a banned substring — usually a short root word or a *word* wildcard in your custom keyword list. AutoMod matches text literally, not by meaning, so "ass" inside "class" triggers the rule. Switch the keyword to a whole-word match or add the clean word to the rule's allow-list.

How do I add an exception to a Discord AutoMod rule?

Open Server Settings → AutoMod, edit the keyword rule, and add the clean word or phrase to the rule's allow-list (exempt words). You can also exempt specific roles and channels from that rule. Allow-lists are per-rule, so add the exception to every rule that could catch the word.

Should I delete AutoMod or just tune it?

Tune it, don't delete it. Native AutoMod is good at blocking scam links and Discord's maintained slur presets server-side and for free. The fix for false positives is narrowing over-broad keywords and whitelisting, not removing protection entirely.

What's the difference between AutoMod and AI moderation?

AutoMod matches a fixed list of keywords and patterns, so it can't tell banter from a real threat. Context-aware AI moderation reads the intent of a message and adapts per channel, which removes most keyword false positives. Many servers run both — AutoMod for hard link blocking, AI for judgment calls.

Is there a free way to get AI moderation on Discord?

Yes. PeakBot includes context-aware AI moderation as one of its free features, with no time limit and no trial period. You can add it alongside Discord's native AutoMod and review everything from a single analytics dashboard.