P:01
Spam pattern gate
Bot networks, generic 'check my profile 🔥', emoji-only spam, and obvious follow-bait are rejected up front. The AI never sees them.
Pre-model filter
Feature · Spam + safety
A safety layer before the AI runs.
Spam, duplicate questions, and policy-risky comments are caught at the gate. The model only generates replies for comments that are worth answering — saving you tokens and stopping weird outputs before they happen.
The six gates
Layered defence, per comment.
Every incoming comment passes through six checks before the AI ever sees it. Each gate is opt-out, transparent, and visible in your activity log — nothing is silently discarded.
P:02
Policy gate
Comments touching topics on your refer-to-you list (refunds, medical, custom collabs) skip the AI entirely and route to your queue.
Topic-aware
P:03
Confidence gate
If the AI's confidence in its draft is too low, it doesn't send. The comment routes to manual review with a flag explaining why.
Quality gate
P:04
Similar-question cache
Recurring questions reuse approved replies — same answer, same voice, lower latency, lower cost. The cache learns from every comment you approve.
Cost-aware
P:05
Rate limiter
Per-account send rate limits ensure replies stay within Meta's expected pacing. Even on viral days, your account behaves like a human, not a bot.
API-safe
P:06
Skip log
Every skipped comment shows up in your activity log with a reason. Nothing is invisible — you can always promote a skipped comment to manual review.
Transparent
Why it matters
The cheapest reply is the one you didn't generate.
Most AI reply tools push every comment to the model. That's expensive — and worse, it produces a reply for things that should never have been replied to.
ReplyMagic's gates are deliberately conservative. The spam gate catches the obvious. The policy gate catches what you said is sensitive. The cache catches repetition. The confidence gate catches the AI's own uncertainty.
What's left is the comments that are genuinely worth a reply. The AI runs faster, costs less, and never says anything that you'd want to delete.
FAQ
Common questions about safety gates.
Why filter before the AI? +
Two reasons. First, it saves you money — every skipped comment is a generation request that didn't run. Second, it stops the AI from saying anything weird about a comment that wasn't worth answering in the first place.
How much does this actually save? +
Depends on your spam-to-real ratio, but typical accounts skip 15–30% of comments at the spam gate. On a 3,000-reply month, that's 450–900 generations avoided — real money on the AI side, plus less noise in your activity log.
What if a real comment gets gated by accident? +
Skipped comments are visible in your activity log with the gate that caught them and the reason. You can promote any skipped comment back into the reply queue manually, and the system learns from your override.
Does the cache reduce reply quality? +
No — the cache only kicks in for genuinely similar questions, and replies are re-personalised to the new commenter (their language, their handle, the specific phrasing they used). It's the same answer, never the same surface text.