> ## Documentation Index
> Fetch the complete documentation index at: https://docs.daftplug.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Content moderation

> Blocking toxic comments before they are submitted.

**Generatify > Settings > Content Moderation**

## AI comment moderation

<ParamField path="AI Toxicity Checker" type="toggle" default="off">
  Checks a comment as it is being written and blocks submission if it detects toxic content: threats, insults, obscenity, identity-based hate, or sexually explicit language.

  The commenter is told the comment cannot be submitted and can edit it. Nothing reaches your moderation queue, and nothing reaches your database.
</ParamField>

## How it works

The model runs in the commenter's own browser. That has three consequences worth knowing.

<Columns cols={2}>
  <Card title="No cost, no provider" icon="circle-check">
    No API key and no per-comment charge, however many comments you get.
  </Card>

  <Card title="No content leaves the visitor's device" icon="lock">
    Comments are not sent to a third party to be scored.
  </Card>

  <Card title="A first-load download" icon="download">
    The model is fetched the first time a commenter reaches a page with comments. It is cached afterwards.
  </Card>

  <Card title="Client-side only" icon="triangle-exclamation">
    Anything that posts comments without running your JavaScript is unaffected.
  </Card>
</Columns>

<Warning>
  This is a usability feature, not a security control. It stops a real person writing something abusive in the heat of the moment. It does not stop a spam bot posting directly to `wp-comments-post.php`, because that never runs the check. Keep your existing spam protection.
</Warning>

## What it does and does not catch

It detects the categories the model was trained on, listed above. It does not detect spam, off-topic comments, link dropping, or politely worded bad faith.

<Note>
  Detection is in English. Toxic comments in other languages will mostly pass through.
</Note>

Expect occasional false positives. Quoted abuse, reclaimed slurs, and frank discussion of a sensitive topic can all read as toxic to a classifier. Because blocking happens before submission, a false positive means a legitimate commenter simply cannot post, and will not tell you. On a site with heated but legitimate discussion, weigh that against what you gain.

## Pairing it with the moderation queue

The checker and WordPress's own moderation settings solve different halves of the problem.

|                          | Toxicity checker  | WordPress moderation         |
| ------------------------ | ----------------- | ---------------------------- |
| When                     | Before submission | After submission             |
| Stops                    | Abusive language  | Anything matching your rules |
| Visible to the commenter | Yes, immediately  | No                           |
| Works without JavaScript | No                | Yes                          |

Run both. Under **Settings > Discussion**, keep your keyword list and your held-for-review rules in place; the toxicity checker reduces what reaches them rather than replacing them.