X shadowban check: what X's own tool actually shows

#Quick answer

The honest X shadowban check in 2026 is X's own one. Open x.com/i/under_the_hood while logged in and it lists every visibility label applied to your account and your posts in the previous month, what each label does to reach, and how many of your posts got it. Your account needs to be over a year old with at least 10 posts in that month. The report is aggregate, so it will tell you that a post was limited and not which post.

Last updated: September 2026

#TL;DR

Shadowban is a user word, not a platform state. X calls the underlying thing visibility limiting, documents it on its enforcement page, and has labelled affected posts since April 2023. Since August 2026 you can read your own labels in a monthly report, and since September 2026 that report also covers posts downranked or withheld because a government demanded it, naming the country. The report has real gaps: no post IDs, no daily breakdown, no memory of labels that were lifted. Third-party checkers cover none of that ground either, and most reach drops are ranking rather than enforcement.

#What X does instead of shadowbanning

Start with the vocabulary, because it determines what you can actually check.

X's enforcement page lists limiting post visibility as one of its standard post-level actions. The page is specific about what that means in practice: excluding the post from search results, trends and recommended notifications, removing it from the For you and Following timelines, restricting its discoverability to the author's profile, downranking it in replies, restricting Likes, replies, Reposts, quote posts, bookmarks and pinning, and excluding it from carrying adjacent ads. That is a list of mechanisms, and several of them can run at once.

What people miss is the next paragraph on that same page. Since April 2023, X says, posts identified as violating its rules receive a label informing both the author and viewers that visibility was limited, and the author can appeal that label. So the secret-punishment model that the word shadowban implies stopped matching the documented product three years ago. There is a label. It is meant to be visible to you.

Account-level actions are separate and more obvious: read-only mode, a lock pending phone or email verification, or suspension. If you have hit one of those you already know, and what to do about a suspended X account is a different process from anything in this article.

#Under the hood: the official visibility report

In August 2026, alongside open-sourcing the For you ranking code, X shipped a page that reads your labels back to you.

You reach it at x.com/i/under_the_hood, logged in, on web or in the app. The report covers the previous calendar month and splits into post labels and account labels. Each label comes with its name, a short description of what it means, a line on its effect on visibility, and the count and percentage of your posts that received it. Engadget and TechCrunch both put the eligibility gates at an account older than one year and at least 10 posts published in the month being reported, which rules out new accounts and quiet ones.

Version one handed you a raw JSON file and nothing else, which drew enough complaints that X added a rendered view on the page. Both still exist: read it on screen, or download the JSON if you want to diff month over month. The label names in that file are machine-style constants such as SPAM_HIGH_RECALL, which is worth knowing before you open it and wonder why nothing is written in English.

This is the first time the platform has handed authors the moderation signal directly rather than leaving them to infer it from impressions. Pair it with your native X analytics and you can separate a labelled month from a merely quiet one, which no amount of staring at an impressions graph will do on its own.

#The legal withholding disclosure, added September 2026

On 19 September 2026 X expanded the report to cover a category that used to be invisible unless you happened to travel.

Under the hood now shows whether your account or your posts had visibility limited because X had to comply with local law, and it names the country that demanded it. X Open Source framed the change as a response to how well the earlier transparency work had landed. Practically, it closes a real blind spot: a post withheld in one jurisdiction looks completely normal from anywhere else, so an author in Berlin could have a post suppressed in another market for months without a single signal.

X's country withheld content policy is the backdrop here, and it is more generous than most people assume. Withholding is limited to the jurisdiction that issued the valid legal demand. X says it will promptly notify affected users and indicate the withholding in the product unless it is prohibited from doing so, giving a sealed court order as the example, and that the notice typically identifies both the content and the origin of the request. Appeals go back through the notification email for a legal demand, or through the Help Center when the withholding came from a local-law report.

#Four things the report still will not tell you

An open issue on X's own algorithm repository, filed on 20 September 2026, is the clearest inventory of the gaps, and it is written by someone who read the payload.

It does not name the post. The report says one of your 412 posts was labelled. It does not say which, which means you cannot open that post, read the label and appeal it. The issue puts it sharply: a count without an identity is a notification, not transparency.

It has no timeline. Day-level data exists in the pipeline and gets collapsed into a single monthly integer before you see it. A label applied on the 3rd and a label applied on the 29th are the same number to you, so you cannot line a label up against the post that caused it.

Lifted labels vanish. A label that was applied and then removed reads identically to one that was never applied at all. If you appealed successfully last month, the report gives you no record of it.

Ranking is out of scope. The report covers enforcement labels. It says nothing about how the recommendation model scored your posts, which is where most reach variance actually lives. The open-sourced For you code is the place to look for that, and it is a different question entirely.

#The manual checks worth running, and the ones that lie

If you are outside the eligibility gates, or you want to confirm a specific post, there are two checks with real signal and several with none.

Worth doing: open the post in a logged-out browser window. If it loads on your profile but does not appear in search for a distinctive phrase from its own text after a few hours, that is consistent with a search-exclusion label. Also open your profile logged out and confirm your replies appear under the posts you replied to, since reply downranking is one of the listed mechanisms.

Not worth doing: comparing this week's impressions to last week's and calling the difference a ban. The For you model ranks every candidate post against everything else competing for that slot, so topic, format, timing and what your audience happened to be doing all move impressions by more than most enforcement actions do. Engagement rate is the more stable read, because it normalises out the reach swing.

Third-party shadowban checkers sit in an awkward middle. They automate the logged-out search test, which is a genuine check, and then present the result as a verdict on your whole account, which it is not. They cannot see account labels, cannot see a withholding in a country they are not querying from, and produce false positives on posts that search has simply not indexed yet. Use one as a search test if you like. Do not read it as a diagnosis.

#If your reach dropped and the report is clean

This is the common case, and it is worth saying plainly: a clean report plus falling reach usually means the problem is your posting, not a punishment.

Failure modes repeat. Posting into the afternoon trough rather than either daily peak. Front-loading a post with a link so the preview eats the hook. Replying only to large accounts, which reciprocate least. Running the same format for six weeks until the audience stops registering it. None of those produce a label, and all of them produce a graph that looks like suppression to someone who is looking for suppression.

One behaviour does cross into enforcement territory: volume. X publishes daily limits on posts, replies, DMs and follows, and sustained bursts against those ceilings are exactly the pattern its platform manipulation policy targets. Staying well inside them is the cheapest insurance available.

#Where automation sits in this

Running an engagement tool changes your exposure here, and pretending otherwise would be dishonest.

Automating replies through a browser is a gray area under X's automation rules rather than a sanctioned path, and the sanctioned path is the API. Our own testing hit a verification challenge doing exactly this, which we wrote up in full at the honest risk. X-Autopilot runs on your own Mac, from your residential IP, in your own browser, with the session cookie never leaving the machine and no delegate account involved. That is a smaller detection surface than a cloud scheduler acting from a shared IP pool. It is not an absence of risk, and no tool can promise you one.

The longer answer, including what the ban waves actually correlated with, is in will I get shadowbanned for using automation.

#The short version

Check the official report first. If it names a label, appeal on the post and fix the behaviour that earned it. If it names a country, read the withholding notice and decide whether to challenge the demand. If it comes back clean and your reach is still down, stop running shadowban checks and start reading how the X algorithm ranks a post, because that is where the answer is.