AI Companionship

SpicyChat

How a global leader in AI roleplay moderates 380M conversations every week — with 95%+ precision and 100% AIGC output coverage.

380M

Weekly Moderation Volume

95%+

Avg. Precision Per Label

100%

AIGC Output Coverage

The Customer

A dual-engine platform
built on open interaction.

SpicyChat is a global leader in AI open dialogue and character role-playing applications. Its core business model is driven by a dual-engine of UGC and AIGC — enabling users to create and customize AI characters for immersive, unrestricted interaction.

Given its high user stickiness and open-ended nature, the platform demands rigorous compliance and safety standards for every AI-generated output — without compromising the creative freedom that defines the product.

At a Glance

IndustryAI Companionship

Content ModelUGC + AIGC dual-engine

ChannelsMobile App + Web

DeepCleer ProductsText Moderation · Real-Time AIGC Audit

The Challenge

Open-ended AI raises
complex safety stakes.

As global users surged, SpicyChat faced safety challenges far more complex than traditional platforms.

CHALLENGE 01

High-Risk AIGC Outputs

Because AI generates dialogue dynamically, outputs can drift into severe policy violations that no static keyword filter can anticipate — requiring real-time interception before content ever reaches the user.

Underage ThemesNon-ConsensualBestialityIncest

CHALLENGE 02

Disparate Channel Standards

Mobile platforms like Google Play enforce extremely strict regulations, while the Web version requires more flexible moderation — necessitating differentiated control strategies that a one-size-fits-all policy can't deliver.

App Store RulesWeb Flexibility

The Solution

A closed-loop strategy
tuned to every channel.

Full-Volume Real-Time Inspection

DeepCleer implemented a closed-loop "Model Output → Instant AI Audit → Violation Interception" workflow — ensuring every AI dialogue is strictly filtered before it reaches the user. No generated output ships unchecked.

Differentiated Strategy Configuration

Distinct moderation benchmarks were configured for App and Web channels — meeting stringent app store regulations on mobile while preserving business flexibility on the Web, all from a single integration.

Granular Label Management

Precise interception logic was established for sensitive categories. By using configurable labels to distinguish subtle scenarios — fictional anthropomorphism vs. actual illegal acts — the platform significantly improved moderation accuracy without over-blocking legitimate creative content.