The short answer: yes, if you tell the AI what to keep as clearly as what to hide, and if every decision shows its reason and can be undone. A keyword filter can't tell a worried customer writing "is this a scam?" from a scammer. An AI that reads the whole comment can, but you still need to review what it hides, because no moderation system gets every call right.
Why moderation tools hide complaints
- Keyword lists. Words like "scam", "fraud", "refund" or "worst" appear in spam and in real complaints. A list hides both.
- Negativity filters. Some tools hide comments that sound negative. Most complaints sound negative.
- Toxicity scores. Angry customers swear. A model trained to flag rude language flags them too.
- Rules written for hiding only. If the instructions only describe what to remove, anything close to the line gets removed.
Same words, different comments
| Comment | Keyword or negativity filter | Rule that reads context |
|---|---|---|
| "Paid three weeks ago, still nothing. Is this a scam?" | Hidden ("scam") | Kept, reply drafted |
| "Congrats! You won. Claim your prize at our support page" with a link | Missed if no listed word | Hidden: scam link |
| "Worst customer service ever, nobody answers emails" | Hidden (negative) | Kept, reply drafted |
| "The staff at your Main Street store are idiots and so are you" | Hidden (insult) | Your call: hide the insult, reply in a DM |
| "Refund me or I'll report you, order 48213, call me on 0901 123 456" | Hidden ("refund") | Hidden for privacy, reply in a DM |
How to write rules that protect complaints
Write the rules the way you would brief a new community manager. Say what to keep, then what to hide, then what to do in between.
- Keep: "Keep complaints and criticism of our products, delivery, prices and service visible, even when they are angry or use strong language."
- Hide: "Hide scam links, fake giveaways, accounts pretending to be us, spam, hate speech and threats."
- In between: "If a complaint includes an order number, phone number or address, hide it to protect the customer and draft a reply asking them to message us."
- Tie-breaker: "If you are not sure whether a comment is a complaint or spam, keep it and flag it for review."
The tie-breaker matters most. A spam comment left up for an hour costs little. A real complaint hidden for good costs a customer, and often the screenshot they post somewhere else.
Check what the AI hides
You can't measure mistakes you can't see. Pick a tool that shows why each comment was hidden, then:
- Skim the hidden comments once a day for the first two weeks, then once a week.
- Unhide any real complaint, reply to it, and note which rule caught it.
- Fix the rule, not just the comment, so the same mistake doesn't repeat.
- Once a week, also look at a sample of comments the AI kept, to catch spam that slipped through.
If unhiding becomes rare, the rules are working. If you keep unhiding the same kind of comment, the rule that hides it is too broad.
What to do with a complaint instead of hiding it
- Reply in public within a few hours, so other readers see you respond.
- Acknowledge the problem in one sentence before you explain anything.
- Move order details and personal data to a private message.
- Come back to the thread when it is solved. "Sorted, thanks for your patience" under a complaint is good publicity.
Hiding a complaint rarely makes it disappear. On Facebook the person who wrote a hidden comment and their friends can still see it, and the customer usually notices that nobody else reacts.
When hiding a complaint is the right call
- It contains personal data, such as phone numbers, addresses or order details.
- It includes threats, hate speech or harassment of a named employee.
- It is part of a coordinated spam wave copied across many posts.
In each case, reply privately so the customer still gets an answer.
How Social Moderator AI handles complaints
Social Moderator AI was built around this problem. You write rules in plain English, including what to keep. Each action in the inbox says which rule it applied, for example "Kept: complaint about delivery, reply drafted" or "Hidden: link to a fake support page". Any decision can be undone in one click, and the reply drafts follow your brand voice and your own FAQ. It still makes mistakes, which is exactly why the reason is shown. See also our guide to reducing toxic comments without silencing real debate.
Frequently asked questions
Is it OK to hide negative comments on my own page?
The platforms let page owners hide comments, but hiding honest criticism tends to damage trust once people notice. Hide spam, scams, abuse and personal data, and answer the rest.
How accurate is AI comment moderation?
Accuracy rests on the rules you give it and the comments your page gets, so ignore any headline accuracy figure. Test a tool on your own comments for two weeks and count how often you have to unhide something.
Will customers know their comment was hidden?
Usually not right away, because they can still see their own comment. They do notice when nobody replies or reacts, and some post again or complain elsewhere.
Should the AI reply to complaints automatically?
Let it draft replies to complaints and have a person approve them. Automatic replies work well for simple questions, but an upset customer deserves a human check before anything is posted.