Moderation actions in Stream fall into two categories:
- AI-Automated Actions
- Applied instantly by the AI based on policies configured by Admins.
- Moderator-Driven Actions
- Applied by you, the moderator, after reviewing flagged content in the queue.
As a moderator, you don’t configure automated rules, but you need to understand what the AI has already done before you make your own decision.
AI-Automated Actions
When harmful content is detected, the AI takes an immediate action based on the rules Admins set. These actions determine whether the content is visible, hidden, or blocked before you ever see it.
Flag
Flagging marks content as potentially harmful and sends it to the moderation queue for human review. The content remains visible to everyone.
When to Use
- When the violation is ambiguous and needs judgment.
- For low-severity or borderline cases.
- To monitor emerging harms before stricter rules are applied.
A common mistake is to overuse the flagging option, sending too much content into the queue for review. This can quickly overwhelm your team and slow down response times. Admins should only use Flag when there’s genuine uncertainty, not as a fallback for every case.
This keeps the queue manageable and ensures moderator attention stays focused on the most important content.
Block
Blocking immediately removes content. Neither the sender nor other users can see it and the decision is logged in the queue.
When to Use
- For zero-tolerance harms where exposure causes direct risk.
- For violations where caution is better than allowing harmful content through.
A Block should be reserved for high or critical harms where community safety clearly outweighs leniency. Blocking ensures that severe violations are removed immediately, protecting other users from exposure to harmful content.
Shadowblock
Shadowblocking means the sender sees their message as if it were delivered, but no one else does.
When to Use
- For trolls or spammers testing moderation boundaries.
- To avoid giving bad actors feedback that their content is being removed.
By hiding content from everyone except the sender, adversarial users who thrive on attention are slowed down, making their efforts feel ignored and ineffective.
However, a common mistake is applying shadowblock to sensitive harms such as self-harm disclosures. In these cases, concealment may prevent users from receiving the support they need. Shadowblock should only be used where hiding the content won’t cause further harm to the individual.
Bounce and Flag
Bounce and flag bounces the message back to the sender with a prompt to revise. If the user resends it unchanged, it is flagged in the queue. If the user resends it revised, it goes through and isn’t flagged in the queue.
When to Use
- For teaching moments, nudging users to improve behavior.
- For borderline violations like profanity or mild insults.
Bounce and Flag works best as an educational tool. Rejecting a message and prompting the user to revise it, giving users a chance to learn and follow community guidelines in real time. It’s an effective way to encourage self-correction without escalating to stronger enforcement.
Bounce and Block
Like Bounce and Flag, it sends the message back to the sender with a prompt to revise it. If the user resends it without changes, it is blocked entirely. If the user resends it changed, it goes through and isn’t added to the queue.
When to Use
- For clear violations where content should not be allowed, but self-correction is encouraged.
- Good for spam, slurs, or prohibited links.
Bounce and Block gives users a single opportunity to correct themselves before their content is permanently removed. It’s especially useful for “no-go” harms where correction is possible, such as spam links or profanity. If the user complies and revises their message, it passes through as normal; if not, the system blocks it entirely.
Mask and Flag
Mask and flag filters or replaces only the offending part of a message (e.g., profanity → ****). The message stays visible with edits logged.
When to Use
- For profanity, offensive slurs, phone numbers, or spammy links.
- In high-volume chats where removing whole messages would break flow.
Mask and Flag is best used for minor, common violations where you want to keep the conversation flowing but still signal that moderation has occurred. By masking only the offending word or phrase (e.g., profanity) and leaving the rest of the message intact, it strikes a balance between enforcement and user experience.
Custom Severity
Custom severity allows admins to map harms to severity levels (Low, Medium, High, Critical), each tied to a different action.
When to Use
- For harms that vary in intensity.
- To triage the queue by urgency.
When reviewing flagged text content, severity levels should always guide your triage. Critical harms should be reviewed and acted on immediately, while lower-severity items like mild insults or casual spam can wait until higher-priority cases are cleared. Consistency across the moderation team is key, ensuring that severity tags are applied and interpreted the same way by everyone.
As a moderator, your role is to review automated decisions, confirm their accuracy, and override them when necessary.
Moderator-Driven Actions
Once flagged items appear in the queue, it’s your responsibility to decide their final outcome. These actions give you direct control over what happens next.
Mark Reviewed
Mark Reviewed clears the flag on an item and records that a moderator has reviewed and approved the content. The message stays visible in the channel, and no further action is taken.
Use when: Content was correctly actioned by AI or filters, or when the violation is so minor that further enforcement wouldn’t be necessary.
Delete Content
Delete removes the flagged content from the channel while leaving the user account active. This keeps the conversation safe without escalating against the user.
Use when: The message clearly violates community guidelines but doesn’t rise to the level of a ban. Examples include off-topic spam, inappropriate memes, or a one-off insult.
24-Hour Channel Ban
This action temporarily bans a user from posting in a specific channel for 24 hours. They retain access to other parts of the app or community.
Use when: A user is repeatedly disruptive in one channel (e.g., spamming a livestream chat) but isn’t causing problems elsewhere. This allows moderation to cool down the situation without over-penalizing the user.
24-Hour Global Ban
This ban removes a user from all channels across the platform for 24 hours. It’s a short-term, platform-wide enforcement designed to stop widespread disruption.
Use when: A user is spamming or harassing across multiple channels, but their behavior may be temporary or impulsive. The ban acts as a timeout rather than a permanent removal.
Channel Ban (Permanent)
A permanent channel ban blocks the user from a specific channel indefinitely, while still allowing them to participate in others.
Use when: A user has repeatedly violated rules in the same
channel and shows no signs of changing behavior. This is common in communities where users are disruptive in one topic area but behave appropriately elsewhere.
Permanent Global Ban
The most severe action, a permanent global ban removes the user from all channels and prevents them from participating in the community entirely.
Use when: A user engages in severe harms (e.g., threats, child safety violations, hate speech) or repeated violations across channels, showing they cannot safely remain on the platform.
Best Practices for Actions
- Match the action to the severity of the harm.
- Escalate step by step, start with temporary bans before going permanent, unless it’s a critical case.
- Use notes to explain decisions, especially overrides.
- Stay consistent with other moderators to avoid uneven enforcement.
Now that you know the full range of AI and moderator actions, let’s look at how to work faster and smarter in the queue, using bulk actions, keyboard shortcuts, and notes to handle high-volume moderation without losing accuracy.