> ## Documentation Index > Fetch the complete documentation index at: https://docs.tightknit.ai/llms.txt > Use this file to discover all available pages before exploring further. # AI Moderation > Have AI check new messages for spam, code of conduct violations and criteria you write, then decide what happens with automation rules. > [!NOTE] > > AI moderation is available to pilot participants only and is off by default. Contact [Support](mailto:support@tightknit.ai) if you'd like to join the pilot. AI moderation checks every new message in your community as it arrives. For each message, AI says how sure it is that the message is spam, breaks your code of conduct, or meets a criterion you wrote. Only admins see those results. Nothing happens to a message until an [automation rule](#automation-rules) acts on them. Message text has personal information removed before it is sent to the model, the same way as other [AI features](https://docs.tightknit.ai/ai/overview.md). ## Where to configure this In Studio, open **[Moderation](https://studio.tightknit.ai/-/moderation/settings)** under **Engagement** in the sidebar. It holds the on/off switch, the built-in checks and your custom criteria. **[Automations](https://studio.tightknit.ai/-/moderation/automations)** holds the rules that act on the results, and is linked from the top of the Moderation page. Admins can always open both pages. To let someone else manage moderation, give them the **Moderation Manager** [role](https://docs.tightknit.ai/studio/access.md#roles). A Moderation Manager can change every setting and rule on these pages except adding **Delete message** to a rule: that needs an admin, because the rule deletes messages as them in Slack. ## Turn AI moderation on or off The **AI Moderation** section has one switch, **Check new messages**. * **On:** new messages are checked as they arrive. Turning it on never re-checks older messages. * **Off:** no message is checked and no rule runs. Your checks, criteria and rules are kept, and apply again once you turn it back on. Every change to these settings is recorded in the [audit log](https://studio.tightknit.ai/-/workspace/audit). ## Built-in checks Tightknit writes these checks and keeps improving them. Each one is on by default. | Check | Default threshold | What it looks for | | - | - | - | | Spam | 0.80 | Content posted mainly to promote, sell or redirect rather than to take part, such as ads, crypto schemes or copy-pasted promotions | | Code of conduct violation | 0.85 | Messages that break your [community guidelines](https://docs.tightknit.ai/moderation/community-guidelines.md) | Code of conduct is checked against the text of your community guidelines. If your guidelines are only a link to another page, AI has nothing to check messages against, so the check stays off until a Studio admin adds the text on **[Onboarding](https://studio.tightknit.ai/-/onboarding)**. ### Thresholds A threshold is how sure AI must be before a message counts as matching a check, from 0.05 to 1. A threshold of 0.85 means 85% sure. A higher threshold catches fewer messages and makes fewer mistakes. A check's threshold decides when a message counts as flagged for it. It doesn't change your [automation rules](#automation-rules): each rule has its own threshold, so raising the Spam threshold here doesn't make a spam rule act on fewer messages. To change one, open the **⋯** menu on the check and choose **Change threshold**. **Use default** fills in Tightknit's default. Save it and the check keeps following the default, even when a later release changes it. Changing a threshold never re-checks messages. Turn a check off with its switch to stop asking about it. Results it already produced are kept. ## Custom criteria A custom criterion is a yes/no question you write, asked about every new message alongside the built-in checks: for example, *"The message mentions BrightHub, Loopwise or Kindred by name."* A community can have up to 10. * **Add criterion** opens a form for the name, what to look for, and a threshold (0.70 unless you change it). * Describe one condition, phrased positively. A message that meets it gets a yes. * Editing what a criterion looks for saves a new version. Past results keep the version they were judged on, and older messages aren't re-checked. * **Archive** stops a criterion being checked and keeps its results. An archived criterion no longer counts toward the 10. ## Test a message To see what AI moderation would do before you turn a check, criterion or rule on, select **Test** next to **Test a message** on the Moderation page, or **Test rules** on **[Automations](https://studio.tightknit.ai/-/moderation/automations)**. Both open **[Test a message](https://studio.tightknit.ai/-/message-test)**. * **Write a message**: type the text, pick the feed it would be posted in and whether it's a post or a reply. Nothing is saved, and no one is notified. * **Use a real message**: paste a Slack message link or a post URL from your community site. The results also show whether the message was already checked with the current settings, and what a re-run would change. The results list each built-in check and custom criterion with how sure AI was against its threshold, then each rule: * **Matched**: the rule's condition matched, so its actions would run. * **Would match**: the rule matched but is logging only, so it would only be recorded. Matches are also reported this way while AI moderation is off. * **No match**: the rule watches the message's feed, but its condition didn't match. * **Not in this feed**: the rule doesn't watch the message's feed. Rules that are off aren't listed. For a keyword rule, the results show which keywords matched. The Test screen shows only what you're allowed to manage. A Moderation Manager sees moderation results; someone with the **Tags** permission sees [AI tagging](https://docs.tightknit.ai/tags/getting-started.md#testing-ai-auto-tagging) results; an admin sees both from one test. For a real message, **Re-run on this message** checks it again with the current settings and applies the result, running your rules. A re-run only runs what you're allowed to test and is recorded in the audit log. You can run about 30 tests a minute. ## Automation rules A rule decides what happens when a check, a criterion or a keyword matches. Open **[Automations](https://studio.tightknit.ai/-/moderation/automations)** to see every rule, how often each matched in the last 7 days, and a switch for each. **[New rule](https://studio.tightknit.ai/-/moderation/automations/new)** opens the rule editor, and **Browse templates** fills it in from a preset for spam, code of conduct or blocked words. Each rule has three steps: * **When:** an AI check or criterion at a threshold you choose, or a list of keywords. * **Where:** all feeds, or the feeds you pick. * **Then:** one or more actions: **Add tags**, **Flag for review**, **Alert a channel** (a private channel or the Tightknit Hub) and **Delete message**. While you write a rule, the preview checks it against the last 500 messages in its feeds, using results already stored, so you can see what it would have caught before you turn it on. A new rule starts as **Logging only for 7 days**: it records what it would have done without doing it. You can also save it **Off** as a draft, or turn it straight **On**. > [!WARNING] > > **Delete message** acts as you in Slack, so only a Studio admin who is a Slack workspace owner or admin can add it. We recommend a threshold of 90% or higher, and starting the rule as logging only. ## Review flagged messages **[Review queue](https://studio.tightknit.ai/-/moderation/review)** lists every message that needs a person's decision: messages an AI check or criterion scored at or above its threshold, and messages a rule flagged with **Flag for review**. Open it from the top of the Moderation page or from **Automations**. Each message shows the check that flagged it, the check's version, the guidelines version for code of conduct, the score, and what rules already did to it. For each message you can: * **Delete message**: deletes it in Slack with your own Slack account, and from your community site. The first time, Studio asks you to connect your Slack account so Tightknit can delete on your behalf. Slack only lets workspace admins and owners delete other people's messages, and a post with replies can't be deleted until its replies are removed. * **Message the author**: sends them a Slack direct message from the Tightknit bot. The message stays in the queue. * **Deactivate user**: walks you through deactivating them on Slack's own member management page, since Slack only lets admins and owners deactivate members there. * **Dismiss**: closes the review and leaves the message as it is. If the author edits it, AI checks it again. Every action is recorded in the audit log. If one fails, the message shows what went wrong and what to do about it. Messages you dismissed or deleted move to **Resolved**. ## Related topics - [Getting Started with Moderation](https://docs.tightknit.ai/moderation/getting-started.md) - [Getting Started with Tags](https://docs.tightknit.ai/tags/getting-started.md) - [Community Guidelines](https://docs.tightknit.ai/moderation/community-guidelines.md) - [Access the Studio](https://docs.tightknit.ai/studio/access.md) - [MCP Server](https://docs.tightknit.ai/integrations/mcp.md) This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.