YouTubeFuncionesPreciosBlog
Empezar gratis
← Volver al blog

Detección de discurso de odio en comentarios

24/5/2026

Beyond keyword lists

Keyword filters miss context, sarcasm, and multilingual hate speech. SafeComments uses GPT-4o-mini with language-specific prompts.

Confidence thresholds

Comments classified below 60% confidence go to the manual review queue — never auto-deleted without certainty.

Armenian and diaspora content

Our Armenian prompt handles Unicode script, Latin transliteration, and Russian-Armenian code-switching. Read the Armenian guide.