Comments moderated across 3,400+ active publisher sites since 2019.
Source: Comment Castle, Inc. annual operating report, 2024.
The Product
For five years, Comment Castle has been the fortified home for the comments sections that publishers cannot afford to lose. This is the long read on what is actually inside the walls — chapter by chapter.
Chapter Outline
Read in order, or jump straight to the chapter your team cares about. Every module ships inside the same moderation engine, behind the same SOC 2 Type II boundary.
The outer wall. Sub-180ms decisions on every incoming comment, before it ever reaches a moderator's queue. Reported block rate of 99.2% across enterprise deployments in 2024.
Read the chapter →Our patented scoring model, trained on 41 million labeled samples, scoring every comment on a calibrated 0–100 scale with 96.4% precision against brand-safe baselines.
Read the chapter →Badges, reputation, threading, and lightweight moderation prompts that lift comment quality without strangling conversation. Built with former community managers, not just engineers.
Read the chapter →A single console for editorial leads: live queues, toxicity trend lines, repeat-offender graphs, exports for legal and compliance. The reporting room, in one tab.
Read the chapter →One-click connectors for WordPress, Ghost, Drupal, and Discourse — and a documented REST/Webhook surface used by 600+ engineering teams to wire moderation into custom pipelines.
Read the chapter →By the Numbers
Comments moderated across 3,400+ active publisher sites since 2019.
Source: Comment Castle, Inc. annual operating report, 2024.
Average moderation cost reduction within the first 90 days.
Audited across 280 customer migrations in 2023–2024.
Average moderation latency, end-to-end, on a published comment.
Independent benchmarks, CommunityTools Review, Q2 2024.
Public uptime over the trailing twelve months.
Live status page, commentcastle.com/status.
Chapter II · The Engine
Toxicity Score v3 is the patented model at the center of Comment Castle. Every incoming comment is scored on a 0–100 scale, returned with a calibrated confidence band, and routed according to rules your editorial team defines — not rules our engineers hard-coded in 2019.
The model was trained on 41 million labeled comment samples, drawn from English, Spanish, French, German, Portuguese, and Japanese publisher communities, and validated against brand-safe baselines to a measured 96.4% precision. It does not promise perfection — no moderation system does — but it gives moderators a reliable first read, and it does so in well under 180 milliseconds.
In plain terms: a comment arrives, the engine scores it, and three things happen in parallel. Spam is blocked at the gate. Likely-toxic comments land in a moderator's queue, sorted by score. Likely-safe comments pass through, optionally enriched with engagement prompts or reputation rewards. The moderator never has to scroll past the obvious cases to find the genuinely difficult ones.
The same engine ships with on-device GDPR pseudonymization in production — the first moderation vendor to do so at this scale — and the entire stack is SOC 2 Type II and ISO 27001 certified. Compliance is not a bolt-on; it is part of the foundation the scoring runs on.
Chapter III · Growth
Defense is half the job. The other half is making sure the people who bother to comment in the first place feel like the conversation is worth returning to. Comment Castle's engagement layer was designed by former community managers who have actually closed comment sections at 2 a.m.
Lightweight reputation, threaded replies, contributor badges, opt-in slow-mode, and prompt-engineered reply suggestions that nudge readers past the first three words. None of it is bolted onto a separate product. It runs on the same engine that scored the comment in the first place, so a low-toxicity thread and a high-reputation reader get treated differently than a brand-new account with a single suspicious link.
Chapter V · Extensibility
If your publishing stack is WordPress, Ghost, Drupal, or Discourse, the integration is one click. If it is something more custom — a paywall, a native app, a homegrown CMS — the documented REST and Webhook surface is the same one our largest customers use.
Every endpoint is versioned, every change is logged in a public changelog, and every payload includes the same Toxicity Score and confidence band that the moderator console shows. You are not integrating against a watered-down public API. You are integrating against the same engine that processes 1.2 billion comments a year.
The Next Step
No slideware. Pick three threads from your own publication before the call, and a Comment Castle solutions engineer will walk you through exactly how the engine would have scored them — what would have been blocked, what would have been queued, and what would have passed through untouched.
Trusted by publishers reaching a combined 180M monthly unique readers. Winner of the 2024 SaaS Awards "AI Innovation in Media."