Twitter (X) shadowban: what it is, how to check, and how to fix it

A shadowban is reach suppression you are not told about: your content stays up, but X shows it to fewer people, hides it from search, or buries your replies. The word "shadowban" is yours, not X's. X's term is visibility filtering, and its stated philosophy is "Freedom of Speech, Not Reach," which it defines as "restricting the reach of [posts] that violate our policies by making the content less discoverable."

That is not a leak or a theory. X documents the exact moves on its enforcement page, under the heading "Limiting post visibility." In its own words, it can do any of these:

• exclude the post from search results, trends, and recommended notifications
• remove the post from the For You and Following timelines
• restrict the post's discoverability to the author's profile
• downrank the post in replies
• restrict likes, replies, reposts, quotes, bookmarks, sharing, and editing

The phrase traces back to Elon Musk's first weeks running the company, when he posted in November 2022 that "freedom of speech" would not mean "freedom of reach" and that negative content would be "max deboosted." X formalized it as an enforcement philosophy on April 17, 2023. The honest wrinkle, and the one that makes shadowbans so maddening to diagnose: since that April 2023 update, X attaches a visible label to posts it limits for breaking a rule, and lets you appeal. The quieter kind, where a spam model decides your account looks automated, comes with no label at all. More on that gap below.

"Shadowban" is an umbrella for four fairly different things. The four-part taxonomy you see everywhere was inherited from an old checker site, not from X, so it is community vocabulary rather than official terms. Each one still maps cleanly onto a move X describes on its own enforcement page:

The one to know in 2026 is reply deboost. If you reply at any volume, this is the version that bites, because X collapses replies it scores as low-quality behind a "Show more replies" link or the longer "Show additional replies, including those that may contain offensive content" fold. Users have reported for years, including a wave of complaints in 2026, that the offensive-content fold fires on completely benign replies, so seeing your reply buried there is a genuine deboost signal, not proof you said something wrong. If you want the full picture of the ranking system these penalties live inside, we wrote up how the X algorithm works separately. This piece is about the suppression side of it.

Before you diagnose a shadowban, accept the most likely answer: you were not restricted, the post just did not land. This is the step every other guide skips, and the data is brutal about it. A 2024 study in the journal Business & Information Systems Engineering tracked more than a decade of a Reddit support community where people report suspected shadowbans. Of 94,173 suspicions, only about 3.4 percent turned out to be real bans. The other 96.6 percent were false alarms. That figure is from Reddit, not X, so do not read it as an X statistic. The lesson transfers anyway: people massively over-diagnose shadowbans, usually mistaking an ordinary quiet stretch for a penalty.

So before you go hunting, separate the two. A real restriction and a bad week look different:

If you are mostly in the right-hand column, you do not have a shadowban, you have a content or timing problem, and the fix is earning more engagement, not appealing a penalty. If you are in the left-hand column, keep reading.

There is no button on X that says "you are shadowbanned." You triangulate from a few checks, and the order matters because the cheapest, most reliable one comes last. Run these from an account that does not follow you, or while logged out, since the whole point is to see what a stranger sees.

1. Search your handle and a recent post. Open a private window or a second account, search your exact @handle and the exact text of something you posted in the last day. If neither appears, that points to a search ban. The 2026 catch: X now shows logged-out visitors a login wall after a few results, so a fully logged-out check is flaky. Running it from a second logged-in account is the dependable version, and a Google search scoped with site:x.com "your text" reads Google's index instead of fighting the wall.
2. Run a from: search. Search from:yourhandle and switch to the Latest tab. If you have posted recently and nothing shows, that is the cleanest search-ban test there is. The same operator powers most monitoring; the full set is in our advanced search operators reference, and our free search query builder writes the syntax for you.
3. The reply test. Reply to a popular post, then open that thread from an account that does not follow you. If your reply is missing or sits under the "Show more replies" fold, your replies are being deboosted. This is the most reliable manual check in 2026, precisely because it sidesteps the search login wall.
4. Check a post's impressions for a cliff. Tap a recent post and open "View post engagements," or go to analytics.x.com, and compare the last week to the week before. A sudden drop of half or more with no change in how or what you post is the best quantitative signal you can get yourself. Just rule out the ordinary reasons your impressions might be low first, since most drops are a quiet week, not a ban. Treat the exact threshold as a rule of thumb people pass around, not a number X publishes.

The tester tools the top search results still link to are mostly broken, and the most famous one is dead. The original, shadowban.eu, no longer responds at all; its maintainers abandoned a rebuild when X moved its search behind new authenticated endpoints. The successors that replaced it (the yuzurisa, hisubway, and sorsa checkers) still run, but they have been quietly losing tests. The most-used one removed its quality-filter test entirely and says so in its own FAQ, under a heading that reads "Why is the QFD test gone?"

The reason is structural. These tools work by scraping the same public search surface you checked by hand: the from: operator, direct links to your replies, the old quality-filter toggle. Every time X changes how search behaves or pushes more of it behind a login, the scrapers break. Their own documentation is honest about the limit: they can sometimes detect that a specific ban is present, but they cannot prove one is absent. So a green "you're fine" result tells you almost nothing, and a red one is worth confirming by hand. Treat the checkers as a hint, never a verdict.

X gives you exactly two first-party signals, and neither is a dashboard. The first is the per-post label: since April 2023, a post limited for breaking a rule shows a notice to you and to viewers, with a link to submit an appeal. The second is the limited-account message: if your account trips the rules on aggressive following or engagement, X tells you your account's features have been limited and that your posts "may be filtered out of certain places on X, including from search results and notifications."

What it hides is everything else. There is no page that answers "am I shadowbanned." The widely-repeated advice to "check Settings, then Account status" points at a reach-status screen X has never actually shipped, so do not waste time hunting for it. The genuinely opaque case is the silent one: a spam or quality model decides your account looks automated and quietly throttles you, with no label, no message, and no appeal button. Even X often cannot tell you why. Yoel Roth, who ran Twitter's trust and safety team, put it plainly: spam is handled by "hundreds (if not thousands) of different models and heuristics operating in parallel," which makes any single suppression nearly impossible to explain after the fact.

This is also why X's old denials are a word game. Back in 2018 the company wrote "we do not shadow ban," and that is technically true and practically misleading: it does downrank, hide from search, and bury replies, exactly as its current enforcement page describes. It just does not file those actions under the word you use for them.

Almost every real shadowban traces to one root cause: you tripped a spam or manipulation filter, usually by behaving like automation. The triggers are the negative side of the same ranking system that decides everyone's reach. The model predicts how people will react to you, and a stack of negatives (mutes, blocks, reports, "not interested" taps) drags your score down. In the 2023 code X open-sourced, a report carried a weight of −369, the most negative number in the whole file. That figure is a 2023 snapshot and the current values are hidden, but the direction has not changed: negative signals are far more powerful than positive ones.

The behaviors that set those negatives off, roughly in order of how often they do it:

• Looking like a bot. Follow-and-unfollow churn, near-identical replies sprayed across many threads, scheduled bursts, and link spam are the fastest route to a filter. These are the same patterns that give away an actual bot, which we catalog in how to spot a Twitter bot.
• Drawing reports, blocks, and mutes. Whether your content earns them fairly or a brigade manufactures them, those negative signals suppress reach by design.
• Sensitive or rule-violating content. Posts flagged under hateful conduct or sensitive-media rules are the original target of visibility filtering, and the ones most likely to get a visible label.
• Hashtag stuffing. One or two relevant tags are fine. A wall of them reads as spam to the classifier.
• Heavy off-platform linking. There is no explicit link penalty (Musk said as much in April 2025), but the system optimizes for time spent on X, so posts that send people away tend to get less reach as a side effect, not as a punishment.

One myth worth killing while we are here: the claim that "Grok now risk-scores every new post and throttles it until it clears." The open-sourced 2026 code does carry a brand-safety verdict, and a brand-new post does default to a "medium risk" rating before it is reviewed. But read the code and that verdict only decides whether ads can run next to your post, not how far your post travels. It is an advertiser-safety setting that got misread as a reach throttle. The follow-count thresholds you see quoted ("50 follows an hour gets you banned") are folklore in the same vein: useful as a reminder not to go on a follow spree, not numbers X has ever published.

There is no reset button and no support ticket that lifts a shadowban on request. For the silent, algorithmic kind, you remove the trigger and wait it out. For the labeled kind, you appeal. The steps, in order:

1. Stop the behavior that tripped it, completely. If automation, churn, or copy-paste replies got you here, those have to stop, not slow down. The filter is watching a pattern, and the pattern has to break.
2. Delete the specific offending posts, not your history. If three to five posts clearly crossed a line, remove those. Do not mass-delete, because deleting hundreds of posts at once looks automated and can add a flag rather than clear one.
3. Revoke shady connected apps. In Settings, under connected accounts, remove any third-party tool that posts, follows, or likes on your behalf. Those are a common silent trigger.
4. Slow down and act human. Post and reply at a normal human pace for a few days, with real, varied replies. You are rebuilding the model's read of you.
5. Appeal anything with a label. If a specific post carries a visibility-limited label, use the appeal link on it. That is the one case where X gives you a real button to push.

How long does it take? Honestly, X does not say, and anyone who tells you "48 to 72 hours" is repeating a number with no source. A silent suppression lifts when the signals that triggered it age out and your behavior reads as normal again, which is days of consistent, human activity rather than a fixed clock. A labeled, policy-based limit clears when your appeal resolves or the content comes down. Repeat offenses clear slower than first ones.

The durable fix is two-sided: stop looking like the thing the filters are built to catch, and measure your reach in a way a broken tester cannot fake.

On the prevention side, the biggest lever is reply quality. Reply deboost specifically targets low-effort, near-identical, or bait replies, so the same habit that grows an account, replying with something worth reading, is also what keeps you out of the suppressed bucket. Keep automation off the human parts, hold hashtags to one or two, and do not run follow sprees. If you reply at volume from a tool, variety and substance are what separate you from a reply farm in the model's eyes.

On the measurement side, stop relying on a one-shot checker and watch your own trend line instead. The reliable question is not "did a tool flag me," it is "are my real mentions and impressions actually down versus my baseline." Free X analytics covers your own posts, and a saved keyword monitor with bot scoring tracks whether your brand is still surfacing in conversations over time; the applied version is in how to monitor brand mentions. And remember the consumer side of visibility filtering is yours to use on purpose: muting words and switching to the Following tab are the same downranking levers, pointed at your own feed instead of someone else's.