Moderation & safety

Real-time moderation

On-site: each comment is analysed as it is posted and routed accordingly, so nothing toxic surfaces while the team sleeps.
Social platforms: Logora moderates comments on Instagram, YouTube and Facebook, with around 95% handled automatically through a Mistral-based model.

Analysing and acting on user content at the moment it is submitted, rather than during a later batch review, so toxic or non-compliant contributions are caught before they reach the audience.

Updated June 7, 2026
Topic real-time moderation · real time comment moderation · instant moderation

Real-time moderation means analysing a contribution and acting on it the instant it is submitted, instead of letting it sit until a later batch review. The comment, argument or vote is scored before it ever reaches the audience, and the decision (publish, hold, block) happens in the same request.

Why it matters

Toxicity does its damage in seconds, not hours. A hateful reply visible for ten minutes has already been read, screenshotted and shared. Deferred sweeps, where a team reviews yesterday’s comments this morning, leave a window in which the worst content is the most visible content. For a publisher, that window is a trust problem and a legal one : the DSA expects timely action, not next-day cleanup.

Real-time moderation closes the window. Clean contributions go live immediately, so the reader experience stays fluid, while toxic ones are stopped at the door instead of being chased after publication.

How AI makes it possible

Acting in real time at scale is only feasible because the analysis is automated. A toxicity model evaluates the text as it arrives and returns a score and a category in milliseconds. That speed is what lets a publisher auto-approve the clean majority and auto-block the clearly toxic, reserving human attention for the genuinely ambiguous cases. Without AI scoring at submission, real-time moderation collapses into either a slow pre-moderation queue or no moderation at all.