Content moderation

Content moderation is the work of applying a publisher’s rules to the comments, arguments, votes and contributions that readers post. It is not the same as deleting hate speech (that’s a small subset). It is the daily editorial decision of what counts as a valid public contribution under your masthead.

What’s at stake

For a newsroom, moderation is the dial that controls whether the comment section feels like an editorial space or a liability. Three things ride on the quality of that work :

1. Reader trust. Subscribers leave when comments turn toxic. Der Spiegel data shows 21% of subscribers cite the debate space as a reason they stay, but it can swing the other way just as fast.
2. Legal exposure. The DSA, the GDPR, national hate-speech laws and platform-specific obligations all map to moderation. Bad logs = bad audit.
3. Cost. A regional daily can spend €100-300k/year on manual moderation. The AI-first model brings it down by 80%.

Moderation modes

There are four operational modes, in order of friction :

• Post-moderation. Comments publish immediately, the team reviews after. Fast for the reader, exposes the publisher to short windows of bad content.
• Pre-moderation. Every comment is reviewed before going live. Safe, slow, costly.
• Reactive moderation. Comments are public; the team only acts on flagged content. Cheap, but blind to volume effects.
• Hybrid AI + human. AI auto-handles 85% (auto-approve clean, auto-block toxic), team arbitrates the remaining 15%. This is the modern standard for press websites.

The Logora moderation stack

Logora ships a hybrid moderation pipeline by default :

• AI layer : multilingual model (FR, DE, IT, ES, PT, EN) trained on 1M+ comments from European newsrooms. Outputs a toxicity score and a category (hate, spam, off-topic, threat).
• Human queue : the 15% the model is unsure about lands in a dedicated admin view with keyboard shortcuts (3× faster reviews after February 2026 release).
• Editorial policies : configurable per client. A regional weekly does not have the same rules as a national tabloid.
• DSA logging : every decision is journalised with timestamp, moderator, applied rule, reason text. Annual transparency report assembled in minutes.

Related concepts

• AI moderation, the automated layer
• DSA, the EU regulation that frames this work
• Transparency report, what to publish about moderation
• Hate speech detection

See Logora vs Netino for the difference between a moderation platform and a moderation BPO.