Content ratings and moderation · Chazz.ai

How content ratings and moderation works on Chazz.ai. How content is rated, who decides, and what a rating changes for readers and creators. Every piece of content carries one of two ratings: safe, or restricted. The rating controls who can see it, not whether it may exist. How a rating...

How content is rated, who decides, and what a rating changes for readers and creators.

Every piece of content carries one of two ratings: safe, or restricted. The rating controls who can see it, not whether it may exist.

How a rating is decided

Not by the creator alone. Every save runs through screening before it lands:

This means an item self-heals: something wrongly marked restricted goes back to safe on a clean re-save. Source: worker/src/routes/content.ts:327 guardRatingDowngrade.

A moderator can pin a rating. Once pinned, re-saving will not clear it, which is the point. Source: worker/src/routes/admin.ts:192.

When the classifier cannot be reached, a public save is refused rather than allowed through. Private drafts still save. Failing closed is deliberate.

Images are re-fetched and classified. Every image on the item must point at this platform's own storage, and the actual bytes are checked. Anything illegal is refused outright.
All text is screened. Every string in what you saved is normalised against obfuscation and checked.
The verdict can only ratchet up. If the screen finds restricted material, the item becomes restricted regardless of the label you chose. Labelling something safe cannot keep it safe.
Downgrades are re-verified. Moving an already-restricted item back to safe only succeeds when the screen can actually confirm it, including re-checking images the save did not itself carry.

What a rating actually changes

Catalogue listings: A viewer who is not eligible sees only safe items. This is enforced on the server, so it applies to the whole catalogue and not just the visible page.
Opening a single item: A restricted item returns an eligibility error to a viewer who is not verified, except its owner, who can always read their own work.
Images in the gallery: Restricted images are blurred with a badge for viewers who are not eligible.
Marketplace listings: Listing a restricted item forces the listing to that rating.
Comments: A comment inherits the rating of the thing it is on.

What happens during a conversation

The rating on a character does not decide what happens in your chat. Your own account status does.

Steering in character is a deliberate choice: a character finds a reason to change the subject instead of breaking the scene with a refusal message.

Callers on the native mobile apps are always moderated, whatever their account status, because of app-store policy. The full experience is on the website.

Eligible viewer: Replies stream live. An unconditional check for illegal material still runs on both your input and the model output.
Everyone else: The conversation is steered towards safe content in character, rather than producing refusals. Replies are checked before they are shown; anything that fails is displayed as blurred text of the same shape rather than being silently dropped.

Reporting something

Content you think is mis-rated or breaks the rules can be reported from the item itself. See Getting help and reporting content.

Elsewhere on Chazz.ai

This document is the server-rendered form of /docs/content-ratings. The interactive page needs JavaScript, and shows the same content.

The complete documentation of this service is available as one plain-text file at /llms-full.txt. A short summary, with the operator's policy for automated readers, is at /llms.txt. Changes are published as a feed at /rss.xml.

Chazz.ai is operated by Cognitive Industries, ABN 62 794 528 747, Australia. Contact: [email protected].