How to moderate a non-English Discord server

The moderation stack most owners inherit was built for English communities. Preset word lists cover English profanity best. Filter patterns assume English spelling and grammar. Even where a bot's interface is translated into your language, the moderation engine underneath is still a string matcher, and the strings it knows are mostly English.

Meanwhile your community's actual life happens in Bosnian, Spanish, Tagalog, Arabic, or three languages braided together, full of dialect, transliteration, and slang that mutates weekly. To an English-built filter, all of it is invisible: the genuine harassment and the affectionate banter alike sail through as unrecognized strings.

The traditional answer is doing it yourself. Write custom keyword rules in your own language (Discord's native AutoMod accepts custom word lists, though it caps how many keyword rules a server can have, so the lists must be curated rather than exhaustive). Recruit moderators who are native speakers, because only they can hear the difference between a joke and a threat in your language's register. And write your rules in your community's language, not in English boilerplate copied from another server.

This works, and every non-English community should do all three. Its limit is the same one English servers hit, multiplied: word lists match strings, your language's hostility does not always use the words on the list, and your native-speaker mods still have to do all the actual watching and judging themselves. The tooling that lightens the load for English servers mostly does not exist for you.

It is worth being precise about why filters degrade so badly outside English. Living communities do not speak dictionary language: they speak dialect, respellings, transliteration (the same word written three ways depending on keyboard), loanwords, and code-switching mid-sentence. A word list would need an entry for every variant, and the variants change monthly. In practice the list is always behind, so owners either give up on filtering or drown in false positives from innocent near-matches.

The uncomfortable conclusion: for most of the world's Discord communities, the crude filtering layer barely functions, and moderation is carried almost entirely by unpaid native-speaker humans.

The generational shift in AI is that modern language models understand meaning in major world languages natively. Not translated word lists: actual comprehension of what a message means, in the language it was written, slang and register included. That flips the equation for non-English servers, arguably harder than it does for English ones, because non-English servers are the communities string matching failed most.

This is true of our product Staffd (disclosure: we make it): it judges messages by meaning in the language your community actually speaks, learns your server's specific norms (including which rough banter is normal in your culture), and flags genuine problems to your staff, where your native-speaker humans still make every final call. The division of labor stays the same as any server: AI does the watching, your people do the judging. The difference is that the watching finally works in your language.

The stack for a non-English community in 2026: rules written in your community's language, stating your culture's actual norms. Discord AutoMod on, with a curated custom word list in your language for the unambiguous worst (it blocks before posting, which nothing else can). Native-speaker moderators for judgment and community trust. And AI moderation on top if the watching load is burning your humans out, chosen specifically for whether it genuinely understands your language, which is worth testing before you trust it: paste it your community's actual banter and see if it can tell the greeting from the threat.