On this page11 sections
The short version
Rule-based moderation means you decide, in plain words, what happens to a comment. The software follows those words exactly, every time, and logs what it did.
Automated moderation splits into two camps. With AI-only moderation, a model decides what to hide or allow from patterns it learned, and you set the sensitivity. With rule-based moderation, you write the conditions and the actions, and the engine carries them out. The answer to “why was this hidden?” is the rule itself: you can read it, change it and test it before it touches a real comment.
When a comment arrives on your Instagram and Facebook posts, if it contains “how much”, “price” or “ship”, then reply with the answer and the product link. A real rule, written the way you write it in Social Tools.
When, If, Then: how a rule engine works
Every rule has the same three parts. When is the trigger, If is the conditions, and Then is the actions. Here is one comment going through them, line by line.
One comment on a Cat Novaa Instagram post, start to finish.
| Part | On this comment | Time |
|---|---|---|
| WhenA comment arrives on a post, or a DM comes in | @dani.r · Instagram postHow much is the linen shirt? Ship to Canada? | 9:02 am |
| IfWords, sentiment, intent, platform, author. All must be true. | Matched two of the rule’s words how muchshipRule: Answer price and link questions | 9:02 am |
| ThenReply, private reply, DM, hide, delete, like, send it to you, or a reply drafted by AI | Yes, we ship to Canada. Price and sizes are in your DMs: cmtly.com/linen-shirt | 9:02 am |
| Result | 1 click1 click on cmtly.com/linen-shirt, the reply’s own link | 9:06 am |
The last line is what makes a rule worth writing. Each reply can carry its own tracked link, so you see what the answer led to: a click. And every line is logged: the comment, the rule and branch that acted, what it did, when, and who wrote the rule. How we count a result.
First match wins
Rules are checked in priority order, top to bottom. The first one whose conditions all match acts, and the rules below it don’t run for that comment. Inside a rule, its branches are checked the same way. So put narrow rules above broad ones: spam first, then specific questions, then catch-alls.
Cat Novaa’s four rules, and the comment above.
| # | Rule | @dani.r’s comment |
|---|---|---|
| 1 | Hide spam and abuseLive | No match |
| 2 | Answer price and link questionsLive | Match: replied |
| 3 | Send unhappy customers to youLive | Not checked |
| 4 | DM the link on “LINK”Watch mode | Not checked |
| One rule acted. The rest stayed quiet. | 1 reply |
A rule in watch mode still counts as the match. If it’s first in line, it records what it would have done, and the live rules below it stay quiet too. That keeps a test from colliding with the rules you already trust.
Test it, then watch it
The biggest advantage of rules: you can test any rule on a sample comment before it goes live. Type a comment in, or pick a built-in example, and see exactly which branch would act and why, condition by condition. Tune it until it’s right. No AI-only tool can do this, because its decisions can’t be predicted or replayed. Start your free trial to try the rule builder.
Then let it watch. Any rule can run in watch mode, and starters begin there: it runs on your real comments and records what it would have sent, without sending anything. When the log reads right, switch it to Live.
The watch log for DM the link on “LINK”, this week.
| Comment | Would have | When |
|---|---|---|
| @lea.mLINK | Sent a DM with cmtly.com/linen-shirt | Mon 9:12 |
| @sam.kthis colour is unreal | Nothing: no match | Mon 9:40 |
| 23 would send this week | Sent for real | 0 |
Why rules beat AI alone, for teams that need control
-
You know exactly why a decision was made
When an AI hides a comment, the best you get is “negative sentiment detected.” With a rule you see: matched the words
giveawayandDM me, so branch 2 of “Spam detection” hid it, under a rule Sarah wrote on 15 June. That’s the difference between guessing and knowing. -
No false positives from generic AI
AI models are trained on general data. They don’t know that “cancer” is a research term for a medical nonprofit, or that “kill it” means “great job” in your community. With rules, you control the vocabulary: your rules, your context, your decisions.
-
Test before you trust
Try a rule on a sample comment, then let it watch your real ones in watch mode. Nothing goes live until you say so.
-
Prove it to clients and regulators
When a client asks “why was this comment hidden?”, or a regulator asks how you moderate, a rule is the answer. An AI-only tool gives you a confidence score. For agencies managing brand safety and for regulated industries, that difference matters.
-
AI still helps. It just doesn’t decide.
Rule-based doesn’t mean no AI. Sentiment and intent are AI-scored conditions you can use in a rule. You write “if sentiment is negative and the comment contains a link, then hide it”. The AI scores the sentiment; the rule makes the call.
Rule-based vs AI-only, line by line
AI-only catches new patterns with less setup. Rules give you control and a record. An AI-only tool also runs the same model for everyone, so your comments get the same treatment as anyone else’s.
| Line | Rule-based | AI-only |
|---|---|---|
| Decision transparency | Yes: Traceable to the rule that acted | No: “Negative sentiment detected” |
| Customization | Yes: Brand-specific rules | Partly: A sensitivity slider |
| Testing before it goes live | Yes: Try it on a sample comment, then watch mode | No: Deploy and hope |
| Audit trail | Yes: Per decision, per rule | Partly: Aggregate analytics only |
| Catches new kinds of attack | Partly: Needs a rule update | Yes: Pattern recognition |
| Setup time | Partly: Define your rules first | Yes: Turn it on and it works |
| More than one platform | Yes: Same rules on every platform | Partly: Varies with each platform’s AI |
| Client reporting | Yes: Shows what was moderated and why | No: Shows volume, not reasoning |
When should you use rule-based moderation?
Rules are best when
- You need to prove moderation decisions to clients or regulators
- Your brand has its own terms that generic AI gets wrong
- You moderate on more than one platform and want the same rules on each
- You’re an agency with several clients, each with its own policy
- You want to test before it goes live
- You need an audit trail for compliance, clients or internal review
AI-only may be enough when
- You moderate a single brand with generic needs
- You don’t need to explain decisions, only cut the volume
- Speed of setup matters more than precise control
- You’re a solo creator with one account and no compliance requirements
When a rule is the wrong tool
Some comments need a person. “Ordered 2 weeks ago and nothing yet, order #4471” is not a question a template should answer. Write a rule that sends it to you, as the starter Send unhappy customers to you does: it lands in Needs you with nothing sent. You answer it yourself, and the rules keep that queue short.
A good setup answers the routine questions, hides the spam and hands you the rest. At Cat Novaa, that left 9 of 547 comments for a person this week.
How Social Tools implements rule-based moderation
Social Tools is built from the ground up as a rule engine: you write the rules, the engine carries them out, and AI helps where it’s useful.
- Rule builder
- Rules with several branches. Each branch joins its conditions (words, sentiment, intent, platform, author) with AND, and can run several actions: reply, private reply, DM, hide, delete, like, send it to you, or an AI-drafted reply.
- Try it on a comment
- Test a rule on a sample comment before it goes live: pick a built-in example or type your own, and see which branch would act and why, condition by condition.
- Priority order
- Higher rules are checked first, and a rule’s branches top to bottom. First match wins: predictable, debuggable, auditable.
- Per-decision log
- Every action is recorded: which comment, which rule, which branch, who set it up, and when. From comment to decision, for your own review or a client’s.
- One set of rules, every platform
- Write a rule once and it runs on Instagram and Facebook today; TikTok, YouTube, X (Twitter), Reddit and Discord are coming. Platform conditions tune a rule for one of them when you need to.
- AI-assisted, not AI-decided
- Sentiment and intent are available as conditions, and AI can draft a reply. AI gives the signal; your rule makes the decision.
Every rule can be tested before it runs. Start your free trial to try the rule builder yourself. No credit card required.
Try it on a sample comment
Pick a comment and see what the starter rules do with it, line by line, before you connect anything.
What happened to @dani.r’s comment
| Part | On this comment |
|---|---|
| When | @dani.r · Instagram postHow much is the linen shirt? Ship to Canada? |
| If | Matched the starter’s words how muchship Rule: Answer price and link questions |
| Then | Yes, we ship to Canada. Price and sizes are in your DMs: cmtly.com/linen-shirt |
| Result | 1 click On the reply’s own link |
What happened to @_bot_crypto’s comment
| Part | On this comment |
|---|---|
| When | @_bot_crypto · Instagram postDM me to earn $5k/week |
| If | Matched the starter’s words DM meearn Rule: Hide spam and abuse |
| Then | Hidden. No reply sent. |
| Result | Hidden 1 second after it landed. Hiding is free: no automation used. |
What happened to Priya S.’s comment
| Part | On this comment |
|---|---|
| When | Priya S. · Facebook postOrdered 2 weeks ago and nothing yet, order #4471 |
| If | Matched the starter’s condition sentiment: negative Rule: Send unhappy customers to you |
| Then | Sent to Needs you, in your inbox. No reply sent. |
| Result | Needs you For Maya. Waiting 4 min. |