Moderation settings are easy to ruin in both directions. Switch everything on and members collect warnings for ordinary sentences, then leave. Switch nothing on and the spam arrives. Here is an order that works for most chats.
Step 1. Leave the critical categories alone
Scams, phishing, doxxing, extremism, threats, illegal substances, escort ads, the CAS spammer database, joins through shared chat folders and third-party bots being added — all on by default and all strict. Nobody argues about these: they harm members rather than break etiquette.
Step 2. Switch on what matches your topic
Next come the categories that are essential in one community and a nuisance in another. Decide by what your chat is about, not "just in case".
- Advertising and Telegram links — almost always worth it in support chats and topic communities.
- Crypto, currency exchange, gambling — switch on unless you want your chat to become a classifieds board.
- Profanity and insults — the contentious pair. Insults are worth enabling almost everywhere; profanity depends on the culture of the room.
- Links and bot mentions — useful where newcomers arrive constantly, irritating in a developer chat where people trade links all day.
Step 3. Pick an action, not just "on"
Every non-critical category has its own action: nothing (log only), warning, mute, ban, or send for review. A good rule of thumb: start soft and tighten based on what the log actually shows, rather than on what you fear.
Strictness is separate — soft, standard or strict. It decides how confident the bot has to be before it acts. In a lively chat full of slang, standard beats strict.
Step 4. Decide what members see
This is half of how moderation feels from the inside. The bot always posts something in the chat; what you choose is the wording:
- Full — the bot names the category and the reason. Good where transparency matters.
- Custom — your own text instead of the standard one.
Keep it short. A deleted advert does not need a paragraph of explanation under it.
Step 5. Newcomers and sudden arrivals
The first messages of anyone who just joined are checked more strictly — that is where most spam arrives. Raid protection is separate: a sudden spike of joins turns on slow mode, and it turns itself off once the spike is over. You rarely need to touch either, but it helps to know they exist.
Three ready-made setups
Product support chat. Critical categories plus advertising, Telegram links and flooding. Insults set to warn. People arrive with questions; keep the bot's replies minimal.
Topic community. Critical plus advertising, crypto and currency exchange. Profanity by agreement with the members. Full notifications: in a living community, transparency is appreciated.
Large public chat. All of the above plus links from newcomers and bot mentions. Harsher actions: mute instead of warn.
What to do after a week
Open the analytics and look at which categories fire most. Usually one category accounts for eighty per cent of everything, and that is the one worth tuning. The rest can be left as it is.