Twelve rules, every new comment
Facebook and Instagram give a page a word list. Spam, bait and abuse rarely repeat a word, and they arrive in every language at every hour. The rules read meaning as well as pattern, act as the comment lands, and name themselves on every comment they touch.
Free for 7 days, no credit card required
- Meta's official API, no password, nothing installed
- Hide or delete, your choice per rule
- Bulk hide and delete, no cap
- From $29 a month
Seven comments, six rules, one kept
A spam link, a phone number, an insult, a fake support account, a "DM me" pitch and a row of emojis land one after another. Each is read by the rule built for it and leaves with its verdict. The real question passes all twelve and stays, marked for a reply.
Every rule, what it reads, what it does
Per page, saved once, running on every new comment day and night. The labels are the app's own.
The four rules that read meaning
Model-based checks, marked AI in the app. They read the comment in its own language, with the page's context.
- Profanity
Offensive language in any language, including creative spellings and transliterations that a word list misses.
ProfanityAI
- Negativity
Negative comments read in your page's context. A text box takes your guidance in plain words: what counts as negativity here, and what is a real complaint that deserves a reply instead.
NegativityAI
Showcase guidance: "Anything calling the coffee a scam or overpriced, or telling people not to buy. A real complaint about an order is not negativity, it needs a reply."
- Custom topics
Topics you describe in a sentence each, pressed Enter one by one. Anything on the topic is caught, however it is phrased.
Custom topicse.g. People promoting their own service, press Enter
- Impersonators
Fake profiles posing as your page or its support: the name, the wording and the ask are read together, so "Northwind Coffee Support: send your card details" is caught on a page named Northwind Coffee.
ImpersonatorsAI
The seven pattern rules
Deterministic, language-independent, with allow lists where you need exceptions.
- URLs
Every link is caught, shortened or not. Add the domains you allow (your own shop, your own booking page).
URLsAdd domains to allow
- Emails & phone numbers
Email addresses and phone numbers in any format, the classic "call me for a better price".
Emails & phone numbers
- Mentions and hashtags
Every @mention, every #hashtag. Hashtags take an allow list for your own campaign tag.
MentionsHashtagsAdd hashtags to allow
- Images
Comments that are an image or a GIF: meme replies, random photos, counterfeit-ad images.
Images
- Emojis
Comments that are only emojis, however many.
Emojis
- Keywords
Your own word list, matched anywhere in the comment. Also catch single-word comments turns a bare "crypto" into a match too.
KeywordsAlso catch single-word comments
- Hide all comments
Review mode: every new comment is hidden the moment it lands and waits for you in the inbox. Useful for a launch day, a crisis, or a page that gets brigaded.
Hide all commentsHide every new comment until you review it.
Hide or delete, per rule
The action is a segmented control on each rule, so a page can hide negativity and delete impersonators.
- Hide
On Facebook a hidden comment stays visible to its author and their friends; on Instagram to its author only. Everyone else stops seeing it. It can be unhidden.
Hide
- Delete
Removed from the platform. It cannot be brought back, so the page shows a permanent-deletion note whenever any rule is set to delete, and the Bin keeps the record.
DeleteDeleting is permanent. A deleted comment cannot be brought back.
- Where the rules run
Both posts and ads, Ads only, or Posts only, per page. Only the comments the rules check count toward your plan's monthly comments, so a page set to Ads only spends nothing on organic posts. A new workspace is asked once what to protect, and each page starts there.
Apply moderation toBoth posts and adsAds onlyPosts onlyWhat should CommentWatchDog protect?
- Rules run before automations
A comment a rule hides or deletes gets no automatic reply and no DM: moderation reads it first, so a scam link under your reel is gone before a comment-to-DM automation would have answered it. The automation side of the product is on the DM automation page.
Comment was moderated
- The whitelist
Words that make a comment untouchable: a comment containing one is never hidden or deleted by any rule. "decaf" and "refund" on the showcase page, so a refund complaint always reaches a human.
WhitelistComments containing these words are never hidden or deleted.
Test, copy, save
Nothing changes on your pages until you save, and you can try any comment first.
- The Playground
Type any comment and press Test (Enter on a desktop). The answer comes back with the rule that would act: Would be left alone, Would be hidden, Would be deleted. If the AI is unavailable the page says so; live comments are re-checked when it returns.
PlaygroundType a test comment…Would be left aloneWould be hiddenWould be deleted
- Specimens on every rule
The information mark on a rule opens a card with real example comments that the rule catches, built from your page's own name for Impersonators, so you see what "caught" means before you switch it on.
ProfanityImpersonators
- Apply to other pages
Copies this page's complete moderation setup (rules, actions, scope, whitelist) to any of your other pages you tick. It replaces what those pages had, so the dialog says so in red.
Apply to other pages…Apply to {n} pages
- Unsaved changes are guarded
A sticky bar on phones and a leave dialog everywhere: Save and leave, Discard, or Stay.
Unsaved changesSave and leaveDiscardStay
- Every action is recorded
A comment a rule hides or deletes carries the rule's name in the inbox (Hidden · Profanity, Deleted · URL), can be filtered by label, and counts in Insights. Nothing is silent.
Hidden · {filter}Deleted · {filter}Label
The rules at work
Real captures from the showcase account.
When a rule meets the real world
Outages, false catches, overlapping rules and paused plans are handled on purpose.
- The AI is down
- The seven pattern rules keep running. Comments that need a meaning check wait as pending and are checked automatically when the AI is back, in order. The Playground says "AI checks are unavailable right now" instead of guessing.
- A rule catches a real customer
- Unhide from the row (one click, Undo offered), then add the word to the whitelist or give Negativity guidance so it does not repeat. Deleted comments cannot be restored, which is why Delete asks for a deliberate choice per rule.
- A comment matches two rules
- The first rule to match acts and is named on the chip; a whitelisted word wins over every rule.
- You change rules mid-stream
- New settings apply to comments that arrive after you save. Comments already in the inbox keep their state; run Post cleanup to sweep a post with the new rules.
- Your plan is paused
- Moderation pauses while the plan is inactive: arriving comments are collected and marked unchecked, your filters stay saved, and everything starts protecting again the moment the plan is active, backlog included.
- Nobody set rules yet
- A new page starts with two rules on, both set to hide: Profanity and URLs. Nothing is deleted until you choose Delete on a rule, and everything hidden can be unhidden. The specimen cards and the Playground exist so the next switch is an informed one.
- The month's comments run out
- Moderation waits until the period resets or you move up a plan. Comments keep arriving and are marked unchecked, and your automations keep replying; the account page shows where the month stands.
Rules act first; the rest of the product builds on that
What the rules hide lands labelled in the inbox, what they miss is one sweep away, what they catch is counted.
Facts for choosing
The plain answers, one line each, for anyone comparing tools.
| Question | Answer |
|---|---|
| What it is | CommentWatchDog is an AI comment moderation tool that automatically hides and deletes spam and damaging comments on Facebook and Instagram ads and posts. |
| Hide or delete | Your choice, per rule. Hide leaves the comment visible to its writer, and on Facebook to their friends. Delete removes it for everyone. Either happens the moment the comment lands. |
| Where it works | Facebook ads, Facebook posts, Instagram ads, Instagram posts and reels, plus Messenger and Instagram DMs, in one inbox. |
| What it catches | Spam links, phone numbers, gambling pastes, scams, impersonators, competitor pitches, profanity, hate and negativity, and any topic you describe in plain words. The rules read what a comment means, not only the words in it. |
| Languages | The rules read every language. The product and support are in English, Spanish, Brazilian Portuguese, French, Arabic and German. |
| Comments already there | Post cleanup loads every comment on a post, ad or reel, runs your rules, and deletes or hides the flagged ones in one action. |
| Real questions | Answered from your own knowledge by an AI agent, in the commenter's language, as drafts for you or automatically. |
| Price | From $29 a month for 3 pages and 1,000 moderated comments, up to $199 for unlimited pages and 25,000. Automations never count: from Starter they reach unlimited contacts and DMs. A year costs ten months. Smaller regional plans from $9 in some countries. |
| Free trial | 7 days, every feature, no credit card. If you do nothing after, moderation pauses and nothing is charged. |
| Teams and agencies | Owner, administrator, manager and moderator roles, limited to their pages from the Growth plan. Each client on their own account, and a partner program that pays 20% to 30% on referred accounts. |
| Safety | Runs on Meta's official API and passed Meta's App Review. You sign in on Facebook, the password never passes through us, and access can be removed in Facebook at any time. Data is stored in the European Union. |
| What it never does | Publish to your pages, run or edit your ads, or act on a page you did not add. |
| Who makes it | GrowMeOrganic LLC, Sheridan, Wyoming. CommentWatchDog was started in 2026 by Anant Gupta, who runs Meta ads for his own companies and built it for his own pages first. |
| Support | support@commentwatchdog.com and the chat on this site, in the six languages above. |
The same facts as text for assistants: commentwatchdog.com/llms.txt. Everything on this page describes the product as it runs today.
Questions about the rules
Short answers before you switch one on.
Twelve, per page: Profanity, Negativity, Custom topics and Impersonators read the meaning (AI, any language, creative spellings included); URLs, Emails & phone numbers, Mentions, Hashtags, Images, Emojis and Keywords are pattern rules; Hide all comments keeps every new comment out of sight until you review it. Each one is a switch with its own action, Hide or Delete.
Hiding leaves the comment visible to its author (and their friends on Facebook) and removes it for everyone else; it can be undone. Deleting removes it from the platform for good; Meta cannot bring it back, so the app confirms and keeps a record in the Bin.
Yes. The meaning rules read any language, including transliterations and creative spellings, and the pattern rules do not depend on language at all. The showcase account runs comments in English, Spanish, Portuguese, French, Arabic and German through the same rules.
Yes. The Playground at the bottom of the rules page takes any text and returns what would happen right now, with the rule that would act: Would be left alone, Would be hidden, or Would be deleted.
Your choice per page: Both posts and ads, Ads only, or Posts only. Ads include comments on boosted posts and dark posts your ad account runs from that page.
A whitelist protects words per page: a comment containing a whitelisted word is never hidden or deleted by any rule, and the URL and hashtag rules take allow lists (your own domain, your own tag). Negativity takes plain-language guidance about what counts and what does not on your page.
The pattern rules keep running. Comments that need a meaning check wait and are checked automatically when the AI is back; the Playground says so in the meantime.
Rules are per page, because a coffee shop's Instagram and its wholesale Facebook Page do not read the same. Apply to other pages copies one page's whole setup to any of the others in one step.
Every plan runs the seven pattern rules, Hide all comments, and Profanity and Negativity with their defaults, in any language, on posts and ads. From Starter you write your own AI rules: guidance for Negativity in your own words, Custom topics and Impersonators. A comment counts toward the month's comments only when the rules check it.