Case Study

Gaming Client

Production game chat moderation for one live multiplayer gaming client. Real flag rates from millions of messages in a kids-oriented environment.

What production chat moderation looks like

This is production game-chat moderation for one paying client running live multiplayer chat with younger audiences. The data below shows what the production text API actually flags at scale.

Game chat is not comment-section toxicity. Most flags in this window are not just slurs. Suggestive/sexual, kids-adjacent, and PII sit next to harassment.

Paxmod delivered 87% toxicity reduction in gaming communities through real-time AI moderation with COPPA safeguards.

Messages scanned

12,863,245

30 days ending 13 September 2026

Flag rate

1.04%

134,231 flagged messages

Largest category

Suggestive/sexual

33% of flags

Category mix (30 days ending 13 September 2026)

Rolled-up taxonomy for one gaming client. Raw model reason strings are not published.

Suggestive/sexual

44,542 flags

33.2%

Kids-adjacent

17,508 flags

13%

PII

15,268 flags

11.4%

Other

13,305 flags

9.9%

Harassment

12,507 flags

9.3%

Profanity

9,910 flags

7.4%

Self-harm

9,802 flags

7.3%

Hate

5,943 flags

4.4%

Threatening

3,486 flags

2.6%

Medical advice

1,956 flags

1.5%

Counts sum to 134,227. Tiny remainder vs 134,231 total flags is unclassified rounding.

Monthly volume

Coverage starts 15 July 2026 for this client.

Jul 2026

7,462,880 scans

84,118 flags

1.13% flag rate

Aug 2026

13,778,872 scans

143,689 flags

1.04% flag rate

Sep 2026 (MTD)

5,296,439 scans

57,887 flags

1.09% flag rate

Weekly flag rate

First week is a ramp. 1.345% is not the steady rate.

Stacked grouping

Of the 134,227 categorized flags.

Harm

Suggestive/sexual, harassment, hate, profanity, self-harm, threatening

86,190

64.2%

Kids-adjacent

Minor safety flags

17,508

13.0%

PII

Personal information

15,268

11.4%

Other

Other + medical advice

15,261

11.4%

Method

One paying gaming client running production text-moderation API calls in live multiplayer chat.

No identities, no message text, no client name published.

A flag is a message the models marked for review under the project policy.

Raw reason strings are rolled into the fixed taxonomy above. Kids-adjacent includes under-12 and minor-safety related flags, not a claim about the player's age.

Flag rate = flagged messages divided by messages scanned in the same window.

Coverage for this client starts 15 July 2026.

Please note: this is not a latency benchmark. Not an accuracy or false-positive claim. Not a prevalence-of-harm-in-all-games claim. This is production traffic for one client, not a census of gaming.

Download data

CSV exports of this case study's tables.

Download CSV (Category mix)Download CSV (Monthly series)

API-first trust and safety for gaming chat

Paxmod delivers real-time AI moderation with COPPA safeguards and live-ops workflows. Learn more about our gaming chat moderation API.

Gaming Chat Moderation API Talk to Trust & Safety