Most Telegram anti-spam bots are built around one assumption: spammers
write messages. So we match text against patterns, run it through NLP,
check for suspicious links. But what happens when the spammer sends no
text at all?
Reaction spam is exactly that. A bot joins your group silently, then
floods every post with 🤡, 18+, and gambling emojis — harming your
channel's reputation without triggering a single keyword filter.
Our approach at ModerAI: instead of analyzing message content, we score
behavioral signals. Things like — does this user react but never comment?
Are they reacting from outside the group via channel post comments? Does
their bio contain obfuscated text patterns? Each signal contributes to a
spam probability score. Cross a threshold — you get restricted.
No text needed.
What behavioral signals have you found most reliable for detecting
non-text spam in group chats?
