How TwitchGuard decides a follower isn't real

A follow-bot attack looks impressive for about an hour. Your follower count jumps by a few hundred, the dashboard celebrates, and then the problems start: your analytics are poisoned, Twitch's recommendation signals get noisier, and when the accounts get purged later your count visibly craters in front of your community.

TwitchGuard treats follow-bots as a detection problem, not a cleanup problem. The goal is to catch the burst while it's happening, not to apologize for it afterwards.

No single check decides anything

The core design rule: one suspicious signal is a coincidence, several together are a verdict. TwitchGuard runs its checks as layers, and each layer alone is deliberately weak:

• Account age. Bot farms register accounts in batches. A wave of follows from accounts younger than your threshold is the loudest single signal there is.
• Username patterns. Generated accounts follow generation rules: random letter strings, real-word-plus-numeric-suffix, repeated structures. Humans pick worse, weirder names than scripts do.
• Burst shape. Real follower growth is lumpy but slow. A follow burst has a shape (tight timing, uniform spacing) that organic excitement never produces, even during a raid.
• Cross-channel history. TwitchGuard runs in many channels, and bans are shared across the network. An account that was banned as a bot in someone else's chat gets blocked from yours before it can post a single message.

When enough layers agree, the bot acts on its own. When the signals are ambiguous, it flags instead of banning, because a false positive against a real new fan costs more than a bot slipping through for a few minutes.

Mods can overrule the model

Every automated action lands in Discord as an embed with the full reasoning attached: which checks fired, what the account looked like, and buttons to undo. If a mod thinks the call was wrong, one click reverses it, and the Request Training flow feeds the correction back so the same mistake gets less likely.

That auditability is the part most moderation bots skip. An action you can't explain is an action your mods will stop trusting, and a bot your mods don't trust gets turned off within a month.

What it costs

Nothing. Follow-bot protection, ban-evader detection, and the whole moderation suite are in TwitchGuard's free tier, permanently. The $8/mo Premium tier adds the engagement side (stream announcements, pets, economy, cross-channel portals), not better protection. We don't think safety should be the upsell.

Install TwitchGuard, or read the full feature list first.