Chat moderation.
When the bot is a moderator it can keep chat clean automatically: tunable spam filters, a banned-word list with per-word actions, and a searchable log of everything it did.
This is live chat enforcement (deleting messages, timing people out). It is distinct from the overlay-side filtering that decides what reaches your chat box and text to speech, covered under Moderation.
Chat filters
Automatic spam protection you turn on filter by filter. Each filter has its own escalation ladder, and exempts mods and you by default.
| Filter | Catches |
|---|---|
| Caps | Walls of CAPITAL LETTERS. |
| Symbols | Messages stuffed with symbols. |
| Links | Posted URLs. By default the filter acts on any detected URL. Optionally, turn on a check for typos so sure links act immediately but ambiguous-looking links (a missing space after a period can turn "ok.So anyway" into what reads as a link, the same shape as someone typing "my store is at ok.so" on purpose) are held for judgment before enforcement. |
| Emotes | Emote floods. |
| Paragraphs | Overly long messages. |
| Repetition | Repeated characters or repeated messages. |
| Zalgo | Glitchy combining-character text. |
| One-man spam | One person flooding the same thing. |
| Language | Messages not in the languages you allow. |
Per filter you set an escalation ladder instead of a single action. Step 1 fires the first time someone trips the filter; each step you add after that fires only if they trip the same filter again within a window you set (120 seconds by default), timed from their previous offense. Let that window pass quietly and they are back to step 1 next time. Ladders run up to 5 steps, and once a step is Ban nothing can come after it.
Actions
Five actions to build a ladder from:
| Action | What it does |
|---|---|
| Soft warn (@ only) | The bot replies to the message, naming the rule. Nothing is deleted and no formal Twitch warning goes out, just a nudge. |
| Twitch warn | Issues Twitch's own warning, which the viewer has to acknowledge before they can chat again. |
| Delete message | Removes the offending message. |
| Timeout | Removes the message and times the user out for the length you set. |
| Ban | Removes the message and bans the user. Final: a ladder cannot have a step after it. |
A channel-wide silent mode suppresses the extra chat notice that normally rides alongside delete, timeout and ban, while still enforcing them. Soft warn is exempt from that switch, since the reply is the action, it still posts even with silent mode on.
Soft warn on step 1, then your real action (delete, timeout or ban) on step 2 within 120 seconds. Most first offenses are accidental; save the teeth for the second try.
Filters and banned words you already had configured were converted automatically: your old single action became a one-step ladder, and the old "warn first" behavior became Twitch warn followed by a timeout within an hour. Nothing changes on your channel until you edit a ladder yourself.
Checking links for typos
The links filter has an optional Check likely typos before acting toggle (off by default). When you turn it on, sure links (ones with a scheme like http://, a www. prefix, or a domain slash path) still trigger your escalation ladder instantly. Links that are ambiguous (or look like typos) are held: the system judges whether they are real links or mistakes, the same way it tells a missing-space typo like "ok.So anyway" apart from someone meaning to type "my store is at ok.so", and applies your ladder if they are judged to be real links. Whitelisted domains are left out of that judgment call by call, not the whole message: a message keeps its non-whitelisted candidates for the check, so it is skipped entirely only when every link in it is whitelisted. If a transient problem prevents judgment (error, timeout, or daily limit reached), that message is skipped and not actioned; when the check concludes the link was meant to be shared, the usual action applies retroactively.
If the toggle is off, the filter works exactly as before: every detected URL is actioned immediately per your escalation ladder, with no judgment. Do not confuse a channel where the toggle is off with a transient hiccup while it is on. When the toggle is off, the filter acts on all links. When the toggle is on but a transient error occurs, the action is skipped for that one message.
Banned words
Keep a list of words and phrases the bot acts on in live chat. The whole list shares one escalation ladder by default, same as a chat filter, but any entry can also carry its own action override, labeled Override: always … (skips the ladder) in the editor, that jumps straight to that action every time and never touches the ladder. This is separate from the overlay-display banned-word list, it is about enforcing in chat, not redacting on screen.
Test mode
Turn on mod test mode and a message from you or a mod that would be actioned is left alone, with the bot instead replying to name the rule that would have fired. It lets you tune filters and banned words live without timing anyone out.
Mod log
Every automated and manual moderation action is recorded in the Mod Log: what happened, to whom, why, and when. It is searchable and exportable, so you can review a timeout or settle a dispute after the fact. Ladder enforcement also lands in your dashboard Activity feed as it happens, alongside your other channel events.