How-To Guide
Am I shadowbanned on X? How to check and what to change
X publishes the labels that limit reach and what each one does. Check your own account in five steps, rule out ordinary causes and fix what you can.
By the Superdirector editorial team. X's repository, X's published label list and vendor pages read November 7, 2026; offer details checked November 11, 2026.
At a glance
Am I shadowbanned on X? How to check and what to change
No outside tool can prove it, but you can check for the real thing. X's open-source code has labels on posts and accounts, and most hide a post from recommendations to non-followers while followers still see it. Check for a notice from X, open X's Under the Hood report if you have access, search three of your posts while logged out, and rule out the 48-hour age filter and other ordinary causes. Then fix what the label names and re-test a week later.
- X lists no duration for any label; the 48-hour and 2-to-14-day figures online come from vendors.
- A logged-out search cannot see labels that only hide posts from recommendations.
- Under the Hood is a pilot, so it is not open to every account yet.
X's label list as a matrix: most labels only hide a post from recommendations, while a few change what everyone sees. Original diagram from X's Under the Hood label file, read November 7, 2026.
All 31 labels from X's file (CSV)Reach log for ten posts (CSV)
Make it work for your brand
Find it. Remake it. Post it. Superdirector turns the viral formats in your niche into your brand's carousels, X posts and video scripts, and keeps the calendar full.
What a shadowban on X is, according to X's own code
You cannot prove a shadowban with a third-party checker, but you can check for the real thing. X's open-source code shows no single shadowban switch. It shows labels on posts and accounts, though X says a few files, such as some botmaker rules and model prompts, are not published, and the most common effect is that a post is hidden from recommendations to non-followers while your followers still see it. Check in this order: any notice from X, X's Under the Hood report if you have access, a logged-out search for three of your posts, and the ordinary reasons reach drops. Only X can show you the labels on your account, and it is piloting a tool to do that.
X's README says whether a post can be shown at all is decided by visibility filtering, separately from ranking, using the viewer's own actions such as blocks and mutes and labels that other systems attach to posts and accounts. Some of those rules apply only when the post is a recommendation from an account the viewer does not follow, and the same post is allowed to a follower. That is the pattern people describe when they say their followers see them and nobody else does.
Is it real? A study posted in December 2020 (arXiv abstract) crawled more than 2.5 million Twitter profiles and found Twitter's explanation for visibility limits, that they were bugs, statistically unlikely. In the full text the share of profiles with a limitation ranged from 0.50 percent to 2.34 percent depending on the group sampled. That is Twitter in 2020, not X in 2026, but it shows the practice was measurable before X published its rules.
The labels X publishes and what each one does to reach
X's Under the Hood code holds the list of labels its report can show you, each with a plain meaning and an effect. These are the ones most likely to matter to a brand account; all 31 are in the label sheet, copied from X's label file.
| Label | Applies to | What X says it means | Effect X lists |
|---|---|---|---|
| SPAM_HIGH_RECALL | post | Post detected by automated systems as one that may contain spam. | Post hidden from recommendations to non-followers. |
| SpamHighRecall | account | Account detected by automated systems as likely to post spam. | The account's posts are hidden from recommendations to non-followers. |
| MALICIOUS_URL | post | Post detected by automated systems as likely to contain a link to a malicious site. | Post hidden from recommendations to non-followers. |
| SPAM | post | Post identified by automated systems or in response to a user report as likely to contain content in violation of X's Authenticity policy. | Post is not shown on X. |
| FOSNR_ABUSE_INSULTS | post | Post detected by automated systems or in response to a user report to contain content in violation of X's Abuse and Harassment policy. | Post hidden from recommendations to non-followers. All users can see a label explaining that the post has limited visibility. |
| FOSNR_ABUSE | post | Post detected in response to a user report to contain content in violation of X's Abuse and Harassment policy. | Post discoverability is restricted to the author's profile. All users can see a label explaining that the post has limited visibility. |
| Compromised | account | Account identified by automated systems or in response to a user report as compromised. | The account's posts are hidden from recommendations to non-followers. Password reset required. |
| AbusiveHighRecall | account | Account detected by automated systems may be in violation of X's Authenticity policy. | The account's posts are hidden from recommendations to non-followers. Security challenge required to prevent automated abuse. |
| ReadOnly | account | Account locked to read-only due to violation of X's Terms of Service policy that requires post removal from the platform. | The account can only read content on the platform until the post in violation is removed. The account's posts are hidden from recommendations to non-followers. |
Read the last column first. Most entries say hidden from recommendations to non-followers. A few change what everyone sees. SPAM means the post is not shown on X, and the FOSNR labels put a visible limited-visibility notice on the post.
Most advice about X shadowbans mixes three kinds of evidence: what X publishes, what vendors say, and what a checker shows. This table sorts the common claims.
| Claim you will read | What X publishes | Where the claim comes from |
|---|---|---|
| There is one shadowban switch | Labels on posts and accounts, applied by visibility filtering separately from ranking | Checker sites. Dicloak names four restriction types, and none of them is in X's label list |
| It lifts in 48 to 72 hours | No duration for any label | Unfollr says 48 to 72 hours and Recover says 2 to 14 days, with no X source. Circleboom's reading of X's help page is that a limited account's notice states its own length |
| A checker can tell you | Labels appear in X's own Under the Hood report, a pilot, and the FOSNR labels show on the post itself | Checkers test what logged-out search returns, which is one clue |
| Hashtags or links get you shadowbanned | A MALICIOUS_URL label for links to sites it detects as malicious, and no label for ordinary links or hashtags | Vendor posts, without an X source |
Look for a notice before you test anything
X tells you about some limits. Circleboom says X shows a message at your next login when an account-level limit applies, and X's label list says an account labelled Compromised needs a password reset and one labelled AbusiveHighRecall needs a security challenge. The FOSNR labels are visible to everyone: X's text says all users can see a label explaining that the post has limited visibility.
- Open the suspect post in a private window and look for a limited-visibility label under it.
- Check the email on the account and your notifications for a message from X.
- Sign out and back in, and read any prompt before you tap anything.
- If a notice offers verification or an appeal, use it. For a FOSNR label the report says to appeal from the post.
No notice does not clear you. Some labels are applied by automated systems, so go on to the next step.
Account notices appear at login, while a FOSNR label shows on the post for everyone to see. Original diagram of the notice paths in this guide, from X's label list read November 7, 2026.
Open the Under the Hood report if you have it
With its August 13, 2026 release X launched a transparency tool next to the code. The README says it shows aggregate statistics about the visibility-impacting labels on your account and posts, and that availability will expand over time. X's page describes its report as your posts and account for last month, and during the pilot you must meet an eligibility checklist before you can open it. You open it at x.com/i/jf/under_the_hood.
For each post label the report gives a count such as 3 of 40 posts and a percentage; for each account label it counts days, such as 5 of 31. Each label also gets a What it means text, an Effect on visibility text and a How to appeal text, and it can be downloaded as a JSON file. If both the post labels and the account labels say None in this period, X is telling you no label from its list was on your account that month. That rules out the labels. It does not rule out low reach for ordinary reasons, which is step four.
If you cannot open the report, skip to the next step. Where you can open it, note which labels appear and read their effect in the table above, because that decides what to change.
Run a logged-out search test on three posts
X publishes no test for this, so the method here is ours, and it covers search only, not recommendations. Pick three public posts at least a day old. For each one, copy a distinctive sentence. Sign out of X or open a private window. Search the sentence in quotes, then search from:yourhandle with one distinctive word, and open both Top and Latest. Circleboom notes X's search bar accepts operators such as from:, since: and until: typed directly. Write what you find in the reach log.
| Result | What it means | Next |
|---|---|---|
| Found 3 of 3 | No search-level limit visible. A label that only hides posts from recommendations would still pass this test | Go to step four |
| Found 1 or 2 of 3 | Inconclusive. Search is not a complete index, and Top results prioritize relevance over order | Repeat tomorrow with three other posts and compare found and missing posts |
| Found 0 of 3, but they show on your profile | A possible search limit, or an ordinary cause: a protected account, an unconfirmed email or incomplete setup | Check those settings first, then repeat |
Vendor checkers run a version of the same thing. Circleboom's walk-through has you connect your X account to Circleboom before it tests, and Dicloak names four restrictions, including a ghost ban that combines search limits with invisible replies. None of those four names appears in X's published label list, so treat a checker's verdict as a clue, not a diagnosis.
Six moves in one logged-out path, from picking three posts to writing the result in the reach log. Original diagram of this guide's own search test, which X does not publish.
Rule out the ordinary reasons reach drops
A fall in impressions with no label is the usual case, and X's published ranking code explains a lot of it.
| What X's code does | Effect on a post | How to check |
|---|---|---|
| Drops posts older than 48 hours before scoring | A post stops circulating in For You after two days | Judge every post at 48 hours, not at day five |
| Scores by weighted sum, then repeated-author decay, an out-of-network discount and a new-author boost | Posts close together compete with each other, and strangers see out-of-network posts at a discount | Space posts out and compare like with like |
| Weights predicted probabilities of viewer actions: report -234.0, mute -58.8, not interested -53.07, block -31.2 | A post people in its audience hide or mute can fall without any label | Read the replies, and check whether the topic still fits your audience |
| Comments in the code say reports from bad actors mainly affect recommendations for users similar to those actors | Mass reporting does not suppress a post the same way for everyone | Do not treat one angry thread as a penalty, and check whether the people reacting resemble your audience |
| Drops a viewer's blocked and muted accounts and muted keywords | A post can be missing for some viewers and visible to others | Ask two people who do not follow you |
The README warns that these weights multiply predicted probabilities, not raw counts, so one report does not cancel a fixed number of likes. X publishes no threshold for a drop, so set your own. Log ten posts at 48 hours, take the middle value of their impressions, and compare it with the ten posts before. A drop of more than half that lasts across several topics and post types, with none of the causes above, is a reason to go back to steps one and two. One weak post is not.
Here is a worked example with invented numbers. A founder posts about their app on X and notices replies drying up. There is no notice at login, the Under the Hood report is not open to the account, and the logged-out search finds 3 of 3 posts. The earlier ten posts had a 48-hour median of 1,150 impressions. For the later ten it is 405, about 65 percent lower. In the reach log, the later ten went out in two bursts of five, each inside an hour, and X's ranking code applies a repeated-author decay, which is why posts close together compete with each other.
So the founder changes one thing, the spacing. The next ten posts go out one at a time across the week, and each is judged at 48 hours. If the median comes back, the bursts were the cause. Should it stay low on spaced posts across different topics, the next places to look are the replies and mutes, and after that steps one and two again. Changing the topic, the posting time and the format in the same week would leave no way to tell which change mattered.
A reach drop with no label is usually ordinary, so check the 48-hour filter, post spacing and audience fit before a penalty. Original diagram of the causes in this guide, drawn from X's ranking code read November 7, 2026.
Change what the label points to, then re-test
Change what the label points to. For the spam labels, stop what looks like spam: repeated identical posts or replies, follow churn, paid or traded engagement, and aggressive liking or replying by hand or by automation. Circleboom reports X's authenticity policy names follow churn as engagement spam, and that X holds you responsible for the apps connected to your account, so revoke any you cannot vouch for. For MALICIOUS_URL, remove links to sites that may be unsafe. A Compromised label means resetting your password, and AbusiveHighRecall means completing the security challenge. A FOSNR label is appealed from the post.
Then test again on a fixed schedule. X lists no duration, so do not plan around a vendor's 48 hours.
| When | What to do | What counts as a result |
|---|---|---|
| Day 0 | Make one change, the one the label or your log points to, and write the date in the reach log | A dated note, so later numbers have a before and an after |
| Each new post at 48 hours | Record its impressions at 48 hours, the age after which X's code filters a post out | After ten posts, a new median to compare with the old one |
| Day 7 | Re-run the three logged-out searches on the same posts | Posts found that were missing before, or still missing |
| Next month | Open the Under the Hood report again if you have it | A label gone from the report, which is the only proof it is gone |
If reach is still flat with no label and no notice after that, treat it as a content problem, not a penalty.
Each label maps to the one action that clears it; the same test runs again a week later. Original diagram of the fix steps in this guide, from X's label list read November 7, 2026.
Where Superdirector fits
Superdirector, which publishes this guide, cannot see X's labels and cannot tell you whether your account is limited. Its line is "Autonomous social media growth.*", and the asterisk means AI video and full autopilot are coming soon. On X today it drafts original posts and threads in your voice from the accounts and formats you track. The $199/month plan runs one AI-run social account. Eligible single link-free X text posts can be published to a connected account under its standing posting permission. Threads stay drafts. The $29/month plan works from Claude or ChatGPT and does not post for you. Either way you stay the publisher and label AI content where a platform requires it. See what runs today.