Features
Part of Content moderation policy: the parts worth your attention
What separates the strong content moderation policy examples from the weak
Strong content moderation policy examples from Meta, YouTube, X and TikTok show how clear rules explain an outcome while weak ones leave the user guessing.
What separates the strong content moderation policy examples from the weak comes down to one test: can the person affected work out what happened, and what to do next? A moderation policy is strong when the answer is yes. Meta, YouTube, X, TikTok and Reddit all publish rules. They differ in how much of the outcome they explain.
What to take away
- Meta's Community Standards name the outcome for each rule, and Meta publishes both a quarterly enforcement report and Oversight Board decisions.
- YouTube's Community Guidelines separate what gets removed from what loses monetization or features, and strikes expire after 90 days.
- X's enforcement labels attach a named reason to a post, and its help pages list a time out of 12 hours to 7 days.
- TikTok publishes a warning-strike ladder with fixed expiry and keeps the violation history inside the app.
- Telegram takes abuse reports by email, 4chan moderates anonymously with a short rule list, and Tumblr's December 2018 adult content ban named a category instead of an action.
- A weak policy states a value. A strong one states the action, the reason and the route to challenge it.
Strong examples, by name
Meta Community Standards. Sections such as Hate Speech, Violence and Incitement, and Bullying and Harassment each state the rule and the enforcement, from removal to account restriction. That pairing is what makes a decision checkable.
Meta also publishes the Oversight Board, which reviews selected decisions and publishes its reasoning. The Board's 2021 ruling on Donald Trump's accounts led Meta to set a two-year suspension dated from January 2021.
Meta's quarterly Community Standards Enforcement Report adds action counts and prevalence estimates for hate speech and similar areas. Outside readers can weigh the rule against the numbers, which is the part most policies never provide.
YouTube Community Guidelines and strikes. A first strike brings a one-week upload freeze plus required policy training. A second brings a two-week freeze, and a third within 90 days ends the channel.
Strikes expire 90 days after they land, so the end date exists before the penalty does. Monetization rules sit in the YouTube Partner Program policy. Feature limits, such as comments switched off under videos with minors, sit elsewhere again, so a creator can tell which one applied.
X enforcement options. X labels posts with a reason rather than deleting them silently, and its help pages list the labels, including a notice that hides media behind a click. The same pages set a time out of 12 hours to 7 days and permanent suspension. Community Notes adds context from contributors rather than staff, and X publishes the note-rating algorithm. A label is a middle action made visible, which is the hardest thing to do well.
TikTok Community Guidelines and strikes. TikTok publishes a strike ladder with a fixed expiry and names the violations that trigger each step. The consequence is dated before it arrives, and the in-app violation history shows the strikes on the account. The guidelines also state that some content stays up but is not eligible for the For You feed. A video can sit on a profile and stay invisible to strangers.
Reddit Content Policy. Reddit separates sitewide rules from subreddit rules, and each community states its own in the sidebar. A ban can come from either body. Each subreddit keeps a public mod log of moderator actions, and sitewide suspensions have an appeal route at reddit.com/appeals. The split shows which body made the call.
The band, from lightest to heaviest
Nothing, recorded. The item is flagged, reviewed and left. A record exists that most systems never surface, and repeat assessments often run against records the user cannot see.
A label or interstitial. The item stays and something is attached: a note, a warning screen, a link elsewhere. X's media-behind-a-notice and Meta's false-claim labels both sit here, and Community Notes adds context without removing the post.
Reduced distribution. The item stays and reaches fewer people, sometimes nobody who does not already follow the account. TikTok's For You feed ineligibility and YouTube's demotion of borderline content are the documented versions. This is the most consequential action and the least often quantified.
Removed from discovery surfaces. Reachable by direct link, absent from feeds, search and recommendations.
Age gating or region restriction. Visible to some people and not others, so two users comparing notes will disagree. X withholds posts in countries that order it, the same shape of action with a legal cause.
Monetization or feature restriction. The item or account loses earnings, live features, links or comments. YouTube's limited-features list and advertiser-friendly rules sit in this band.
Removal. The item is taken down, with or without an explanation.
Account-level action. Warnings, temporary restrictions or closure. X's time out runs 12 hours to 7 days, and TikTok ties permanent bans to repeated strikes. What each means and what follows is set out in the outcomes and what each one is.
Weak examples, and what they share
Telegram. Its help pages point abuse reports at the in-app report button or abuse@telegram.org, and say private chats cannot be reviewed. Nothing tells the reporter which rule was applied, and no strike ladder gives a penalty an end date.
4chan. The global rules are a short list, moderator action is anonymous, and there is no per-post explanation or appeal route. The user cannot plan around a penalty that has no stated end.
Tumblr, December 2018. Tumblr announced its adult content ban in December 2018 and enforced it from December 17, 2018, using automated detection and drawing the line in anatomy terms on its help pages. The notice told users a decision had happened, not which rule applied.
Four habits run through those cases.
Weak examples
- Policies that state a value instead of an action. "We remove harmful content" names no threshold and no outcome. The reader cannot tell whether their post was judged harmful or never reviewed.
- Policies with no published strike expiry. The same violation produces different results for different accounts, and the user has nothing to plan against.
- Policies that route every decision to a support form. A form is not a reason. It tells the user a decision exists and nothing about which rule applied.
- Policies that publish rules but not enforcement data. A rule set with no transparency report cannot be checked against outcomes, so nobody outside the company can tell whether the policy is applied evenly.
What each action is actually for
| Action | The problem it solves | The cost it carries |
|---|---|---|
| Record only (Meta's reviewed-and-left items) | Builds a pattern for repeat assessment | Invisible to the person it concerns |
| Label (X media notice, Community Notes) | Adds context without deciding truth | Labels are read as verdicts anyway |
| Reduced distribution (TikTok For You ineligibility, YouTube borderline demotion) | Limits spread without silencing | Unexplainable and unappealable in most systems |
| Discovery removal | Keeps the item available to those who want it | Looks like nothing happened |
| Age or region gating (X country withholding) | Handles material that is fine for some audiences | Produces inconsistent user experiences |
| Feature restriction (YouTube limited features, advertiser-friendly rules) | Targets behavior rather than material | Often the least explained |
| Warning with training (YouTube first strike) | Gives a first-time poster a way back | Depends on the user reading the training |
| Removal | Ends the item | The loudest available action |
| Account action (X time out, 12 hours to 7 days) | Ends or limits the relationship | Consequences reach beyond the item |
Reduced distribution is genuinely proportionate for a large class of borderline material. It is also the action least likely to be explained or appealable. An operator choosing it takes a lighter consequence and a heavier opacity.
What each action solves and costs
Action
- Record only
- Builds repeat pattern
- Label
- Adds context
- Reduced distribution
- Limits spread
- Discovery removal
- Keeps item available
- Age or region gating
- Handles audience fit
- Feature restriction
- Targets behavior
- Removal
- Ends the item
- Account action
- Ends relationship
Problem it solves
- Record only
- Invisible to person
- Label
- Read as verdicts
- Reduced distribution
- Unexplainable, unappealable
- Discovery removal
- Looks like nothing happened
- Age or region gating
- Inconsistent experiences
- Feature restriction
- Least explained
- Removal
- Loudest action
- Account action
- Reaches beyond item
Cost it carries
- Record only
- Label
- Reduced distribution
- Discovery removal
- Age or region gating
- Feature restriction
- Removal
- Account action
Working out what happened to you
Nothing here is a diagnostic, and the honest answer is often that you cannot tell. You can narrow it, in the same evidence-first way described in recording an example so it is worth having.
- Check whether the item is reachable by direct link in a logged-out browser. Present there and absent from feeds points at discovery removal, which is what TikTok's For You feed ineligibility looks like from the outside.
- Ask someone in another country, and someone with a different account age, to look. Differences point at gating, the same shape as a country-level withholding order.
- Look for a notice in every place the service sends them, including an address nobody reads. TikTok, YouTube and X all keep a violation history in the app or on a studio page, and Reddit puts subreddit bans in modmail.
- Compare against your own earlier items, not other people's. Reach varies for reasons unrelated to moderation.
- Keep dated notes. A single quiet week is not evidence; a pattern with dates is the only material an appeal can use. Reddit sends sitewide appeals to reddit.com/appeals, and subreddit bans go to the moderators.
If there is no notice, there may genuinely have been no decision. Attributing every dip to a hidden action wastes the effort better spent where a notice exists.
For anyone designing this band
Two rules make the middle band defensible. Tell the person that something happened, even where you will not say what triggered it. Make every action appealable, not just removals, because unappealable actions accumulate errors nobody hears about.
The reasoning matches what makes a decision explicable, described in a general account of notice, and what makes an appeal to somebody uninvolved worth having, in an explanation of appeal.
Neither describes a requirement binding a private operator. They describe why the structure exists in systems that had to think hard about it.
When a statute or regulator sets the obligation, the clock and wording come from that text, not from your product team. The contract that lets any of this happen is in the agreement between you and the service.
Common questions
Is reduced distribution real or a folk theory?
Meta, YouTube, X and TikTok all describe actions between leaving an item alone and removing it. TikTok's guidelines state outright that some content stays up but cannot reach the For You feed. What is not knowable from outside is whether one caused a particular dip in reach, and that ambiguity is what makes the subject frustrating.
Can I appeal something I was never told about?
In most systems, no, which is the strongest argument against the design. Where a route exists, use it about the item and the rule rather than the reach, since reach is rarely a decision anyone can point at.
Should I delete an item that seems to be limited?
Take a dated copy first. Deleting removes your only evidence of what was published and when, and it does not undo a record held elsewhere.
What makes a moderation policy strong rather than just long?
A named action, a stated reason and a route to challenge it. Meta's Community Standards and YouTube's strike system both do this. Telegram's email route and 4chan's short rule list do not, however many words they run.
