Start with a deliberate AutoMod preset, a short list of channel-specific terms, and a plan for reviewing held messages. Then test ordinary conversation and change the setting responsible for a problem, rather than raising or lowering every filter at once.
For the worked example below, a small gaming channel starts its trial at Level 2. That is an example baseline, not a claim that Level 2 suits every community. The useful result is a setup that catches the messages you need reviewed while letting your normal conversation work.
Before changing anything, record your current preset, any custom category settings, and the terms you have already added. You will need that record if a change makes chat harder to use.
Give moderators a clear description of your chat
Write down the behavior you want before trying to express it through filters. A relaxed game stream might allow mild swearing while still prohibiting harassment, story spoilers, and unsolicited promotion. Those are different decisions; a single “strictness” number cannot explain them to a moderator.
For example, your channel-specific rules could include:
Please keep story spoilers out of chat. Ask before giving gameplay instructions. Do not post unsolicited promotions. Discuss the game and the play, without attacking the people here.
A moderator can use those expectations when reading a held message or a conversation AutoMod did not flag. Whether someone asked for gameplay advice, for instance, depends on what was happening in the stream.
Twitch’s moderation setup guide covers custom Chat Rules that first-time chatters must accept. You can also put a readable version in a Rules panel; the panel setup guide covers adding that information to your channel page. Keep the wording consistent when you update either place.
Choose Twitch AutoMod settings for the right problem
Open the moderation settings in your Creator Dashboard. Separate the controls by what they do:
| Control | Question it addresses | A reason to inspect it |
|---|---|---|
| AutoMod preset and category settings | Which potentially inappropriate messages should be held for review? | A type of ordinary game discussion is repeatedly flagged, or unwanted messages are getting through. |
| Blocked terms | Which specific words or combinations of words need filtering in this channel? | A recurring promotion or channel-specific phrase needs a targeted rule. |
| Permitted terms | Which known harmless terms need an exception to AutoMod’s filtering? | The same legitimate phrase is repeatedly held. |
| Chatter permissions | Which accounts can send messages in the first place? | A viewer cannot submit any chat message because of an entry requirement. |
Twitch documents AutoMod levels from 0, with no AutoMod filtering, to 4, the strongest preset, and allows individual category adjustments. Its AutoMod guide explains these options. Check the actual category settings rather than assuming a preset number describes your whole chat policy.
In our example, the owner records Level 2 as the starting point, checks the category controls against the channel’s tolerance for profanity and harassment, and schedules a short trial. If the same harmless game terminology is held, the next step is to identify the reason and consider a narrow exception. If a broader category is consistently mismatched to the channel, adjust that category and test again.
Keep account-entry requirements separate from message filtering. Twitch’s Chat Verification Settings lets you apply email or phone requirements to all or some chatters, with group exemptions. Changing an AutoMod language category will not resolve a viewer failing a verification requirement.
Build a small term list from actual problems
Use blocked terms precisely
Start with unwanted wording you have actually encountered, rather than importing a long list you have not read. For each entry, write a private note explaining the problem it is meant to address. That makes an accidental match easier to diagnose later.
Be careful with familiar game words that appear inside an unwanted promotion. Blocking a broad word on its own can affect ordinary conversation. A narrower combination may describe the unwanted message better, but you still need to test it against legitimate sentences.
Twitch’s moderation documentation explains an important matching detail: a multiword blocked term matches messages containing all its words, even in a different order. It does not block a message merely because that message contains one of those words.
For a harmless test, imagine temporarily adding sample promo phrase. Check the full combination, a reordered version, and a sentence containing only one word from it. This demonstrates what the rule does without copying an abusive message into your channel. Remove the test entry afterward.
Do not assume that pasting a whole sentence creates an exact-sentence-only rule. The combination of words matters. If a rule catches an unrelated conversation, inspect the matched entry and revise that rule rather than making an exception for every new sentence it catches.
Use permitted terms for a confirmed false positive
First read the reason a message was held. A permitted term is useful when a known harmless word or phrase is the problem; it is not a reason to exempt a whole class of messages without checking their context.
Twitch’s AutoMod guide says permitted terms create exceptions, but another flagged term or a blocked term in the same message can still cause a hold. Adding an innocent item name therefore does not guarantee that every message containing that name will pass.
For the worked gaming channel, suppose a harmless in-game name is repeatedly flagged during ordinary discussion. Review examples, confirm the name is the cause, and add the narrow term you intend to permit. Test it in a normal sentence and keep the rest of your moderation settings intact.
If the cause is your own overly broad blocked entry, fix that entry first. Keeping a bad block rule and adding more exceptions around it makes the list harder to understand.
Make review part of the live workflow
AutoMod holds potentially inappropriate messages for a decision before they appear publicly. It does not make the final contextual judgment for your community. Twitch’s moderation documentation describes the approve-or-deny review process.
Use the AutoMod Queue in Mod View to keep held messages visible while you moderate. Decide who checks it during normal conversation and who covers it when the streamer is concentrating on a game. A message left waiting can interrupt a legitimate viewer’s attempt to join in, even if the filter made a reasonable initial decision.
For each held message, separate the decisions:
- Does this message belong in the conversation under the channel’s rules?
- Is there a recurring filter or term problem worth changing after review?
- Does the viewer’s behavior need any separate moderator action?
Approving one message does not require adding its wording to the permitted list. A phrase may be harmless in this context and unhelpful as a standing exception. Denying a message is also different from deciding to timeout or ban the viewer; Twitch’s AutoMod guide distinguishes holding messages from those actions.
There is a visibility distinction worth checking if an expected message is absent from a moderator’s queue. The guide describes public blocked-term messages as visible for moderator review, while private blocked-term messages are surfaced only to the channel owner. Agree how the owner handles that separate review path.
If you stream without another moderator, plan when you can actually inspect held messages. Test the workload during an ordinary session. Do not treat a long unattended queue as evidence that every message in it needed to be blocked permanently.
Test the experience of an ordinary viewer
Use a willing viewer with ordinary account access for a small set of harmless checks. The owner’s view is not a substitute for the viewer’s experience. Twitch’s channel roles guide documents exemptions for VIPs from several chat controls, including AutoMod, so a VIP test can give you the wrong impression.
Write the expected result before testing. You are checking specific controls, not trying to prove that every possible unwanted message will be caught.
| Test | What to examine |
|---|---|
| A friendly ordinary sentence | Does normal conversation appear as intended? |
| Game terminology your viewers commonly use | If it is held, what reason is shown and does that match your rules? |
| The temporary neutral blocked-term combination | Does the intended blocked-term route work, including the reviewer who should see it? |
| A normal message after changing a narrow permitted term | Did the specific false positive improve without relying on an exempt account? |
| A first-time chatter subject to your entry settings | Can they meet the requirements, and is a verification prompt being mistaken for an AutoMod hold? |
If you also enable Block Hyperlinks, include a harmless link test from an ordinary account. Twitch’s moderation setup guide describes link messages going to review under that setting. A privileged account’s successful post does not establish that a new chatter will have the same experience.
Record the actual outcome and the setting responsible. If a message cannot be submitted at all, inspect participation requirements. If it is held, read the hold reason. If it posts but violates your chat rules, review what a moderator should do and whether a targeted filter change would help.
These are different failure paths. Lowering the overall AutoMod level because one viewer cannot satisfy a chat-entry rule changes the wrong part of the system.
Tune one cause after the trial
Keep a short change log with the date, setting or term, reason, and result. Use descriptions such as “ordinary item name held three times” or “recurring promotion used the same word combination.” Avoid keeping unnecessary copies of personal information or repeating abusive text just to document a decision.
After the trial, choose a change that matches what you observed:
- Repeated harmless holds: Check the matched term or category, then trial a narrow correction.
- Recurring unwanted wording: Add or revise a targeted blocked entry and test ordinary sentences that might also match it.
- Unreviewed messages piling up: Revisit who is watching the queue and when, alongside the controls producing the holds.
- A newcomer cannot send anything: Inspect the applicable chat-entry requirements and exemptions before changing language filters.
Keep the previous value so you can reverse a change that causes new problems. Recheck important settings after changing the games you play, introducing a new recurring phrase, or handing moderation to another person.
The setup is ready for your next stream when ordinary conversation works, reviewers understand the held-message paths, and you can explain why each custom term is there. Continue adjusting from actual examples rather than replacing that knowledge with a higher preset number or a longer copied list.