Social Media Guidelines for Brands: Rules for Ad Comments
read
·

Hide the negative comments. That is the advice paid social teams hear most often, and it is the wrong default for a paid placement. A social media guideline is a written set of rules that tells staff and tools what to say, what to hide, what to escalate, and which platform actions are allowed.
The comment thread under a Facebook, Instagram or TikTok ad gets read by buyers who have not clicked yet. Those buyers treat the thread as part of the creative, so whatever sits under your headline shapes the click the same way your copy does. Social media guidelines get drafted for organic posting and then applied unchanged to paid campaigns, which is where they fail.
Rules alone do not fix this. In Pew Research Center's 2016 survey of U.S. workers, 51% of full- and part-time workers said their employer had rules governing social media use while at work, and 77% said they used social media regardless of whether their employer had a workplace policy. Workplace rules were common. Workers used social media anyway, policy or no policy.
The lesson carries to ad comments. A rule that hides every negative comment strips out your side of the exchange while the criticism itself stays alive in screenshots, replies and group chats. What the next buyer reads instead is silence where your answer belonged. A paid social policy earns its keep the other way around: it decides what the next buyer reads, and it maps every rule to the platform that can enforce it.
Table of Contents
What Social Media Guidelines Must Cover
A working policy covers eight components, and each one exists because a person or a tool has to make a call in seconds. Leave a component out and the decision defaults to whoever is on the account that morning.
The most popular version of this document is a list of banned words. It fails for a simple reason: it can't tell a slur apart from a shipping complaint that happens to contain the same word. A blanket instruction to hide all negative feedback deletes the complaints prospects read before they decide, and it deletes the evidence they need to trust you.
The four questions underneath every component are what to say, what to hide, what to escalate, and which platform actions are permitted.

Component | What It Sets | Example Rule |
|---|---|---|
Brand voice | Tone, vocabulary and claim limits for replies | Replies use plain language and never promise refunds in public |
Comment categories | Which comment types are hidden, answered or escalated | Spam is hidden, product questions get a short accurate reply |
Platform actions | Hide, delete, reply or report rights per channel | Confirm each action in the platform help documentation before writing it into the policy |
Complaint handling | Public acknowledgment and any private follow-up | Acknowledge the specific issue and move order details to a private channel where permitted |
Automation limits | What software may do without a person | No automated replies to safety, legal or harm allegations |
AI disclosure | When synthetic or AI-assisted content must be labeled | Label realistic AI edits where platform rules require it |
Accessibility | Alt text, captions and readable on-screen text | Every video ad has accurate captions before launch |
Review and audit | Sampling, false-positive checks and named owners | A named owner reviews a sample of hidden comments each week |
Facebook, Instagram and TikTok need separate lines under platform actions because the permitted moves differ by channel. "Hide the comment" without a named platform isn't a rule, it's a wish.
The test I use: hand a new team member three real comments and the document. If they have to ask you which platform applies or whether to delete rather than hide, the document isn't finished.
Writing Brand Voice and Reply Standards
Voice is the part of the document people reach for under pressure, so keep it short. Four traits with one sample reply each do more work than a page of adjectives, because a moderator under a live ad can copy the shape, not just the sentiment.
Plain. "Orders leave the warehouse within 48 hours, Monday to Friday."
Specific. "That's the 500ml bottle. The 1L ships separately and is on the same page."
Calm. "You're right that the tracking link went out late. Here's where the parcel is now."
Direct. "Yes. Returns are open for 30 days after delivery."
Then the prohibitions, which matter more than the traits. Replies never promise guaranteed results, never cite a discount that isn't live, never compare to a competitor without published evidence, and never make medical or legal claims. Write these as flat bans with no exception clause. A moderator who has to interpret "usually don't" under a live ad will eventually guess wrong.
The four-part reply
Answer the question in the first sentence.
Add the one product fact that makes the answer useful.
Give the next step: a link, a policy page, or a DM prompt where the platform supports it.
Stop. No sign-off paragraph, no upsell on a complaint.
Channel limits shape step three. Facebook comments can contain clickable links. Instagram comments do not. The same shipping answer becomes a tappable link on Facebook and a "full rates are in the shipping policy linked in bio" line on Instagram. Write the Instagram version first. A reply that works without a tap-through works on every placement, and it forces the reply to carry its own facts instead of outsourcing them to a landing page.
A weak answer to "do you ship to Ireland" reads: "Hi! Yes we do, thanks for asking!" A stronger one reads: "Yes, we ship to Ireland. Orders placed before 2pm leave the same day and arrive in two to three working days." The second answers, evidences and stops. It also survives a screenshot, which matters because a reply under an ad becomes part of the creative every later buyer reads.
Approval rule: one named person signs off on new reply wording before it enters rotation, and the stock reply library gets a review date each month. Copy-paste is how a wrong claim ends up under every ad in the account.
For more on applying one brand voice across placements, see these brand voice examples.
Which Comments to Hide, Reply to or Escalate
The default action depends on the type of comment, not on whether it feels positive. Train moderators on a check order, then let a table settle the rest.
The check order
Is this spam, a scam or harm? If yes, it's a hide or a report and the decision is finished.
Is it a question, a complaint or a threat? Those are three different workflows.
Does answering require order data, a refund decision or a legal judgement? If yes, it goes to a person.
Comment Type | Default Action | Escalate When |
|---|---|---|
Spam or bot promotion | Hide | Pattern repeats across several ads, review the filter |
Scam or fake giveaway | Hide and report where the platform allows | Someone reports losing money or account access |
Product question | Short accurate reply | The answer needs order data or a policy exception |
Objection such as is this legit | Reply with verifiable facts only | The claim cannot be backed by published evidence |
Genuine service complaint | Public acknowledgment, private follow-up where permitted | Refund, safety or legal issue is raised |
Abuse toward staff or identity-based harassment | Hide, report or block under platform rules | Any credible threat is raised |
Credible threat or self-harm mention | No automated reply | Immediate human review, emergency services if danger is imminent |
Misinformation about product safety | Human-reviewed factual reply | Legal or safety claim is disputed |
Hiding is not deleting. The comment stays in place for the person who wrote it, and how visible it remains depends on the platform rather than on your intent. Moderators who believe they deleted something stop watching it.
Threats need their own paragraph in the policy. A credible threat or a mention of self-harm gets no automated touch, no stock reply and no filter decision. It goes to a named person immediately, and emergency services get called when danger is imminent.
Moderation decisions get reversed more often than teams expect. Since the EU's Digital Services Act appeal mechanisms began, users appealed more than 165 million platform content-moderation decisions, and almost 30% were subsequently reversed, per European Commission data. Build the correction path before you need it.
If you're formalising who picks up a disputed comment, this guide to escalation of issues covers ownership by category. Teams building their own tooling rather than buying it can compare content moderation endpoints for developers to see what an API-level filter can and can't return.
Platform Permissions for Facebook, Instagram and TikTok
Meta runs Facebook and Instagram ad comment moderation from the same Ads Manager, but not through identical controls. The documented route is short.
Open Ads Manager.
Select the Ads tab.
Choose the ad.
Open More options.
Select Manage Facebook comments or Manage Instagram comments when the option appears.
On Facebook, the documented actions include liking and replying to comments or replies, plus hiding or reporting inappropriate comments. On Instagram, advertisers can hide or delete. Meta also states that a hidden comment remains visible to the person who posted it while being hidden from everyone else, as set out in Meta's ad comment guidance.
Action | Facebook Ad Comments | Instagram Ad Comments |
|---|---|---|
Hide | Yes, stays visible to the author and their friends | Yes, stays visible only to the author |
Delete | Not listed for standard Facebook ads, and not available on Facebook partnership ads | Yes, including partnership ads |
Reply | Yes | Yes, advertisers can add comments on partnership ads |
Report | Yes | Not documented in the Ads Manager menu |
Partnership ads break the symmetry further. Meta's own documentation for partnership ad comments notes that advertisers cannot delete comments on Facebook partnership ads at this time, while Instagram partnership ads allow hide or delete and allow the advertiser to add comments.
Say an agency runs one creator partnership across both placements and a comment alleges a faulty batch. On Instagram the comment can be removed. On Facebook it can only be hidden from everyone except its author, which means the reply has to carry the whole response. Write the rule per placement, not per platform pair.
TikTok is a separate surface with separate controls, and I wouldn't copy Meta's action list into a TikTok row. Read the current TikTok ad comment documentation and confirm which of hide, delete, reply and report exist for the placement you're buying before the document promises them.
Native versus platform: no moderation tool, ours included, receives permissions the platform doesn't grant. Exerta comment moderation hides and replies to ad comments on Facebook, Instagram and TikTok, and every action still depends on what that channel allows.
For the interface steps on a single comment, this walkthrough on how to hide comments covers the mechanics.
Responding to Genuine Complaints on Paid Ads
Hiding a legitimate complaint doesn't remove it. The author still sees it, and anyone who screenshotted it keeps it. What hiding does remove is your chance to answer in the thread the next buyer is already reading.
The framework worth writing down separates spam, abuse, misinformation, product questions, payment problems and genuine service failures, the split that MediaPost's coverage of complaint handling puts ahead of any list of banned words. A service failure is the one category where a public reply is the asset.

The reply sequence
Acknowledge the specific issue in one sentence. Not "sorry for the inconvenience" but the delay, the wrong item, the missing refund.
State the next step the customer can expect, in plain terms.
Move to order details only through a channel that permits it, and only with the customer's agreement.
Never ask for an order number, address or payment detail in a public comment.
Never promise a refund, a date or an outcome the policy doesn't guarantee.
A weak reply to "ordered 9 days ago, still nothing" reads: "So sorry about the delay! Please DM us your order number." A stronger one reads: "Nine days is longer than our stated four to six. Your order is in the dispatch queue and tracking goes out today. We'll confirm the exact status in DM if you want the details." One names the failure, the other names the fix, and neither promises money.
Acknowledging a complaint isn't conceding an accusation. If the claim is wrong, correct the fact once, calmly, with the policy that supports it, then stop. Arguing in a public thread reads worse than the original complaint.
Define the automation boundary in writing. Allegations of harm, safety incidents, legal threats and repeated failed payments never get an automated reply. Give each category a named owner and a response target the team can hit, then review the misses.
The comment patterns that cause the most trouble on Meta placements are covered in this guide to negative comments on Facebook ads, including the ones that look like abuse and turn out to be a service failure.
Filters, Automation Limits and Review Routines
Native filters catch less than teams assume and more than they expect. On Facebook, a profanity filter and a blocked-word list apply at Page level and extend to ads that Page publishes. Emojis are supported and treated as text characters inside blocked-word rules, per Meta's comment moderation controls. The same controls sit at Instagram account level, so one list needs the same review in both places.
Filters match more than exact spelling. Variations, numbers, symbols and emojis all land in the same ruleset, which is why a list built once and never revisited starts hiding words it was never meant to catch. Review hidden comments and decide, one by one, whether to keep them hidden, unhide them or delete them.
Moderation Assist works forward, not backward. Criteria apply only to comments made after the criteria are created, and the Activity Log records what it hid so an administrator can review it, per Meta's Moderation Assist documentation. Set criteria before a campaign launches. Nothing cleans up the backlog for you.

The review routine
Sample hidden comments weekly. Read what the filter caught, not what you assume it caught.
Track false positives on legitimate criticism. A complaint hidden by a keyword is a buyer you paid to reach and then silenced.
Track false negatives on abuse. Sample the visible thread too, because filters miss misspellings and coded language.
Measure correction time. How long a wrongly hidden comment stays hidden before someone fixes it.
Re-test after every change. New creative, new product, new policy, new language.
NIST's AI Risk Management Framework recommends documenting organizational roles, defining risk tolerance and performance limits, running regular adversarial testing and keeping human review where context demands it, according to the NIST AI RMF. In practice that means testing against comment sets built from sarcasm, slang, misspellings, mixed languages and disputed claims before anything goes live.
A filter tuned to hide every negative word will hide the complaint that told you a product was broken.
For teams whose AI governance sits with a wider group, this overview of AI and social media for executives is a useful framing document. The mechanics of what an automated layer can check are covered in this guide to AI content moderation.
Disclosing AI in Ads and Comment Replies
Most disclosure arguments collapse three separate things into one. Separate them and the rule gets simple. Synthetic creative is generated imagery, video or audio. AI-edited creative is real footage that has been materially changed. AI-assisted replies are comment and DM answers drafted by software.
Only the first two belong to ad-content labeling. Meta applies an AI Info label to content it detects as AI-generated or that advertisers disclose, per Meta's AI content labeling guidance. TikTok has its own disclosure rules for realistic AI-generated or significantly AI-edited content, so check its current policy before each campaign rather than trusting last quarter's summary.
Where the policy usually breaks
Translations. An auto-translated reply is automation, not synthetic media.
Minor edits. Cropping, color correction or a cut that doesn't change a person or a scene isn't a disclosure event.
AI-assisted moderation. Hiding a comment is a moderation action, not content generation.
Automated answers to buyer questions. These need identity clarity, not an AI label.
Replies live under a different rule: identity. A comment reply sits under your ad and reads as part of it, so the next buyer treats whoever is typing as the brand speaking. An automated reply must never present itself as a human representative, and when a buyer asks whether they're talking to a bot, the answer is yes.
Disclose the material synthetic alteration or the simulated identity. Don't label every ordinary automation step as synthetic media, or the label stops meaning anything.
A worked pair makes it concrete. A bot reply signed with a staff member's first name is a violation, because it fabricates a person. A plainly labeled automated answer to a shipping question isn't, because nothing about it pretends to be human. On the creative side, this guide on how to disclose AI video ads walks through what has to appear on the asset itself.
Name one owner for labeling decisions per channel. Facebook, Instagram and TikTok enforce this differently, and a single approver keeps the answer consistent across every ad and every reply posted under it.
Publishing Checks for Disclosures and Accessibility
Accessibility and disclosure belong in the content spec, filled in before the asset is built. Treated as a final review, they get skipped on the ad that ships on a Friday.
The pre-launch list
Alt text on every image. Descriptive, not "product image".
Captions on every video. Accurate, including the offer.
A transcript for anything spoken, linked or attached.
On-screen text readable on a phone, not just on your monitor.
Nothing carried by color alone. Add a label, a pattern or a word.
Charts and comparisons explained in text, because a screenshot of a table is unreadable in six seconds.
Links and buttons with meaningful labels. "Shop the serum", not "click here".
No rapidly flashing content.
The success criteria behind that list come from WCAG 2.2, and the W3C's WCAG 2.2 specification is the reference to cite when someone asks why captions are mandatory rather than nice to have.

On disclosure, the FTC states that social media advertising remains subject to truthful-advertising principles and that an unexpected material connection between an endorser and a marketer must be disclosed clearly and conspicuously. Compensated endorsements and testimonials need an understandable disclosure, not a buried hashtag. Check that it's visible on mobile, that every claim is substantiated, that testimonials carry typical-results context, and that any AI-generated or AI-edited creator content has been approved.
Formatting and permissions vary by channel, so test each placement on its own rather than assuming one publishing rule covers Facebook, Instagram and TikTok. This isn't legal advice, and counsel should confirm the disclosure requirements for your category.
FAQ
Should every negative comment be hidden?
No. Hide spam, scams, abuse and credible threats, and answer genuine complaints and product questions in public where the platform allows it. A hidden complaint still reaches its author, so hiding costs you the reply without removing the criticism.
Is hiding the same as deleting?
No. Hiding leaves the comment in place for the person who wrote it. Deleting removes it, and deletion is only available where the platform documents it for the placement you're buying.
Does a public comment give permission for follow-up messages?
Not by itself. Meta's standard messaging window is generally 24 hours, and for the first message after a Click-to-Messenger or Click-to-Direct ad it may last up to seven days. Confirm the current documentation and the specific entry point before writing any limit into policy, and treat a public comment, or permission for a single private reply, as no basis for ongoing promotional messages.
Who should own the guideline?
One owner for the policy and a separate named owner for escalations. Review the document after every platform rule change and whenever you launch a campaign type it hasn't seen, such as a new placement or a creator partnership.
Other useful guides
This article sets the framework. These pages handle specific comment tasks.
Hide or reply, a moderation playbook for Meta ads gives the triage logic for a single comment when you have seconds to decide.
The 24-hour DM window explained covers messaging limits and what changes after a click-to-message ad.
Negativity is a tax on your ad spend puts the cost of visible negativity in commercial terms.
Get started today
The document is the work. Build it in this order.
Draft the guideline against the eight components, starting with the table in the first section.
Write your own decision table for the comment types you receive, using the check order above.
Set Page-level and account-level filters, then test sample comments containing sarcasm, slang, misspellings and a disputed product claim before any automated layer goes live.
Confirm permissions for each ad placement in current Meta and TikTok help documentation before those actions appear in writing.
Exerta comment moderation handles comments and replies under Facebook, Instagram and TikTok ads, alongside the connected channels available today: Facebook, Instagram, TikTok and your own website. You stay responsible for reviewing and testing how any agent behaves before enabling it on a live channel. The Exerta comment moderation product covers the moderation and reply work, and you can book a demo to walk through your own policy with the team.
Exerta puts AI agents on your ad comments so spam, scams and abuse get hidden before buyers read them, while real product questions get answered in your brand's voice across Facebook, Instagram and TikTok. You keep the disclosure, accessibility and escalation rules in your own document, and the agents work inside them. Start with the placement that carries your worst threads and see how the moderation and reply rules hold up at Exerta.


