AI moderation uses an AI model (OpenAI Moderation) to check posts, blog articles, and events, and, if you choose, replies. It scores content in categories such as harassment, hate, and violence, and can hide or flag content that crosses your thresholds. It works alongside your keyword and spam filter rules, not instead of them.
Who can do this: Owners and admins turn AI moderation on and configure it. Moderators can see whether it's on, read the AI decision log, rate decisions, and review held content. · Where: Web, in the admin console. · Plan: Available on every platform plan. AI moderation doesn't use your plan's AI allowance.
Turn on AI moderation
- Go to Admin → Moderation → Automated Moderation. The switch used to be on the Content Safety page in Settings.
- Under Community-wide settings, turn on Enable AI moderation. It saves right away.
When the switch is off, no content is sent to OpenAI. Content is sent to a third-party service, so review your privacy policy before you turn it on. See AI data privacy and limitations.
Configure AI Classification
The detailed settings live on the AI settings tab of Admin → Moderation → Automated Moderation, in the AI Classification section. You can also click AI settings in Community-wide settings. Only owners and admins see this tab. Everyone else sees a read-only summary under Community-wide settings at the top of the page.
Click Configure and set:
- Mode: Shadow or Enforce (see the next section). New communities start in Shadow.
- Scan: Top-level posts, Replies and comments, or both. Pick at least one. Top-level posts also covers blogs and events, which are scanned when they're created, not when they're edited. Replies generate 5–20× the moderation traffic of top-level posts.
- Thresholds: when content is blocked or flagged, per category.
Click Save changes. Saving here doesn't change the Enable AI moderation switch.
Shadow mode and Enforce mode
| Mode | What happens |
|---|---|
| Shadow | The AI only records scores. Nothing is hidden and nothing is queued. Use it to see how the AI judges your community's content. |
| Enforce | Content over a Block threshold is hidden immediately and sent for review. Content over a Flag threshold stays visible and waits for a moderator's decision. |
When you switch to Enforce, you must tick a confirmation that flagged posts will be withheld from now on. Review shadow-mode results in the AI decision log first.
Set AI moderation thresholds
Each category has a threshold from 0 to 1. A category triggers when the content's score is at or above the threshold, so a lower threshold is stricter.
- Block (reject content): severe categories, such as Sexual content involving minors, Graphic violence, Self-harm instructions, Threatening hate speech, and Violent wrongdoing.
- Flag (send to review queue): broader categories: Hate, Harassment, Sexual content, Violence, Self-harm, and Illicit behavior.
Tick a category to use it and set its value, or click Recommended defaults. Each tier is saved as a whole: categories you leave unticked in a tier no longer trigger. Advanced (JSON) shows the same settings as JSON.
Review content held by AI moderation
In Enforce mode, blocked and flagged content waits in Admin → Moderation → Pending Content, where Reason for review shows AI moderation and the category. Approve it to publish it or reject it to remove it. Blocked content stays hidden until you decide; flagged content stays visible meanwhile. Your decision is also recorded as feedback on the AI's call. For how to review it and what authors see, see Review held and pending content.
Check AI decisions in the AI decision log
The AI decision log tab of Automated Moderation (previously the separate AI Activity page) lists every AI moderation decision. It's a log, not a work queue: acting here doesn't change the content. A line at the top says whether AI moderation is on, what it scans, and how many items are waiting to be scanned or couldn't be scanned.
- Each row shows the Content and its author, what it was Triggered by, the Highest score, the Result, and the Date. Results are Would block or Would flag (shadow, recorded only), Blocked, Flagged, Not enforced, or Passed.
- Filter by mode, result (including Any hit (block or flag)), category, content type (Posts, Replies, Blogs, or Events), and date range.
- Click a row to see the model, thresholds, and Category scores. Old thresholds marks a decision made before you last changed them.
- Rate a decision with thumbs up (Correct call) or thumbs down: Wrong call (overturn) on a hit, Should have been caught on a pass. Click your vote again to withdraw it. Rating needs Resolve moderation items. Decisions you made in Pending Content are already rated.
- Overturn rate shows how often hits shouldn't have been caught, and Missed how often passes should have been. By category shows both per category: a high overturn rate means raise that threshold, and a high missed rate means lower it. Click Adjust thresholds to change them.
- The numbers count decisions Since thresholds were last changed. Switch to All time to include older ones.
Things to know about AI moderation
- AI moderation checks the title and body text of posts and replies when they're created and when they're edited, and of blog articles and events when they're created. It doesn't check images, attachments, or videos.
- Content from community admins and moderators, and from the space's own admins and moderators, isn't sent to AI moderation.
- Keyword, regex, and spam filter rules are set separately. See Set up auto-moderation rules.
- AI moderation doesn't use your plan's AI allowance. See Understand AI usage and credits.
Related articles
Was this guide helpful?