IA AP3K
Sécurité de l'IA et protection des commentaires
Contrôler la façon dont l'IA traite les abus, les critiques, les questions non répliquées et la sollicitation.
Dans cet article
Étape par étape
- AP3K traite le contenu de Instagram comme non fiable et jamais comme des instructions système.
- Choisissez Sauter la réponse ou Supprimer pour les catégories de problème supportées.
- Les questions sans réponse sont sautées plutôt que devinées.
- Gardez les informations sur le produit, le prix, la disponibilité, le remboursement et la politique à jour dans Connaissance.
- Examiner le comportement d'IA avant de l'activer sur une automatisation en direct.
Tap any screenshot to enlarge it. Screens show an example account; your activity and available plan details may differ.
Choose voice and response boundaries
In Behavior, describe the AI role and your brand voice. Choose Fun, Friendly, or Professional as the default tone. In Guardrails & escalation, explain when to avoid guessing and when to offer human help. Keep these instructions consistent with your saved business facts.
Review Comment protection deliberately. Skip reply and Delete are different actions: deleting affects the Instagram comment. The screenshot shows a Delete selection for insults and hate speech as an example, not a required setting. Choose what suits your moderation policy, then save.
Test in Playground before going live
Add at least one knowledge note, save the setup, then ask a realistic customer question in Playground. Try a follow-up and a question your notes do not answer. Check accuracy, tone, and whether the assistant respects your boundaries. Update Knowledge or Behavior and test again if needed.
Playground conversations are saved, but they do not become live business knowledge automatically. Save useful facts explicitly as knowledge. Tests use your monthly AI allowance. When the results are ready, enable the appropriate master switch and AI option in the intended automation, save, and test the live flow.