YouTube comment rules that actually work

A rule is a decision you only make once

Most YouTube comment rules fail in one of two ways. They catch something they should not, or they miss something they should. The ones that hold up are specific, tested before they run, and clear about which of those two failures they are willing to accept.

A rule also moves a decision. Moderating by hand means re-deciding, comment after comment, whether this is spam or a question or someone being rude for sport. A rule makes that call once, when you are calm, so the section runs on a policy instead of on your mood.

The four kinds of rule

Most comment automation comes down to four rule types, and they do different jobs.

Keyword hide. A list of phrases that, when they appear, cause the comment to be hidden on arrival. This is the workhorse. The useful detail is where the matching happens: if it runs locally, before the comment reaches the YouTube API, the rule costs no quota and adds no delay.

Author blocklist. Hiding by channel rather than by message. Keywords catch a sentence. Some spam is a person, and that person rewords the pitch. A rule keyed to the channel ID keeps working when the display name changes, which is exactly what a determined spammer does.

Canned reply. A written answer for the questions you get on every upload. Each rule carries its own template and its own auto-post toggle. Leave the toggle off and the draft waits for approval. Turn it on for the handful of replies you are happy to send unattended.

AI triage. Everything the first three do not catch. A model classifies the comment, and a draft reply is written only when a reply is worth writing. On a channel with a few hundred comments a video, this is what stops the queue from being a pile of unread spam.

Match on phrases, not words

The failure mode of keyword rules is the false positive, and almost every false positive comes from matching a word that means something else in context.

• A rule for "free" hides the viewer asking whether your course is free.
• A rule for "link" hides someone offering you a source you actually want.
• A rule for "crypto" hides the viewer asking whether you cover it.

The fix is phrases. "free crypto" is a far safer rule than "free". "dm me for" catches the pitch without catching a viewer who wants to DM you. "check my channel" catches self-promotion without catching a genuine recommendation.

Two more habits help. Start with contains-matching on phrases, and only widen to patterns or regular expressions once you have watched a rule behave for a few days. And add the specific rather than the general: the phrase your impersonator uses, the misspelling real spam uses to slip past a generic filter. Those are rules only you can write.

Test before it runs

A rule you activate without testing is a rule whose cost you do not know. The way to learn the cost is to run it against real comments first.

A simulator takes a rule and a set of sample comments and shows exactly what the rule would hide, reply to, or pass through. You see the blast radius before a single comment is touched. If the rule catches the scam and leaves the question about your course alone, it is ready. If it catches both, tighten the phrase.

This step is short and it is the difference between a rule that helps and a rule you disable in a week.

Order matters

When more than one rule can match a comment, the order decides the outcome. A comment from a blocked author who also matched a keyword should be handled by the blocklist, not by a reply rule. A scam that happens to contain a question word should be hidden, not answered.

Give the rules that protect the channel a higher priority than the rules that reply to it. Hiding and blocking should win every conflict with drafting. That ordering is what keeps a clever spam message from earning a reply just because it was phrased as a question.

Hiding needs a second signal

A single label is a thin reason to hide something, and hiding is a public act. The comment disappears for everyone except the person who wrote it, and on a busy channel a wrong hide costs a real viewer with no notification either way.

The rule that holds up: require a confident classification and a second, independent signal before a hide fires automatically. A comment that looks like someone sharing contact details is left alone unless it is clearly malicious. When the model is unsure, the comment stays up and is flagged for review instead of hidden. Leaving a doubtful comment visible costs you a scroll. Hiding it wrongly costs you a viewer.

What to automate and what to keep

The rules that hide are safe to run without asking, because hiding is reversible and invisible. A wrong hide can be undone. The rules that reply are different, because a published reply is public and permanent in the way a hidden comment is not.

That asymmetry is the whole design. Let keyword rules and the blocklist run. Let canned replies queue as drafts by default, and opt in a rule to auto-post only when the answer is a fact that never changes. Keep the specific, personal replies for yourself, because those are the engagement you were trying to build.

The automations page walks through each rule type, the simulator, and the approval model in more detail. The features page shows how the rules fit with triage, the review queue, and the activity log.

A good starting set is small: the last ten spam comments you can remember, a phrase that catches your channel's specific scam, and a blocklist entry for anyone who keeps coming back. Add rules as the patterns show up, not all at once.