X Wants You to Check If You've Been Shadowbanned Yourself — But the Answer's Buried in a JSON File
X has added an "under the hood" page in its settings menu, letting a small group of test users download about a month's worth of visibility-limiting records, while also open-sourcing more of its recommendation algorithm code. But what you actually get is a chunk of code, not a straight answer for every single post.
If you've suspected you're being shadowbanned, X is now willing to hand you a file as "proof" — the catch is, that proof looks like a pile of code.
The feature is currently limited to "a small group" of test users, tucked away in a new "under the hood" page inside the app's settings. From there, users can download data tied to their own account and dig through records labeled "visibility-limiting labels," which are supposed to reveal which posts had their reach restricted.
But according to X's own explanation posted on GitHub, what the download actually gives you is "aggregate statistics" — an overview of how often an account or its posts were tagged with visibility-affecting labels over the past month, not a post-by-post breakdown. In other words, you can find out roughly how many times you've been flagged this month, but not what exactly happened to a specific post at the time.
Twitter account @cb_doge (Doge Designer) posted a hands-on demo video, and the downloaded file is a .json — at first glance, just a wall of code. Digging deeper, you can find that one account had two posts tagged "NSFW," which resulted in those posts being hidden from non-followers and underage users.
Alongside this update, X also open-sourced more of the recommendation algorithm code that powers the "For You" timeline. Keith Coleman, X's VP of Product, called it "unprecedented transparency into X's algorithm." The GitHub repository lists a large number of classification labels covering various types of NSFW content, spam, violence, impersonation, and hate speech, along with some more unusual entries — like a "for emergency use only" label reserved for emergency situations, complete with customized prompts.
One notable detail: the list doesn't include any label directly tied to political content, but there is a "civic integrity" flag that restricts a post's visibility to only be viewable on the author's own profile page. According to X's policy, this label is "reserved for major escalation cases," triggered when a user report leads the system to determine that content violates the civic integrity policy.
X says the code released this time was reviewed and stress-tested beforehand by "third-party recommendation system experts," though it didn't explain how those experts were selected or whether the test results would be made public. Elon Musk had previously said that by the end of August, "all code touching the X system will be open-sourced and third-party audited." This release clearly falls short of that bar — the company itself admits it's still withholding some information to prevent people from "gaming the system," and code for systems beyond the "For You" timeline, including the ad system, still hasn't been released. An X spokesperson added that future updates may introduce new labels related to brand safety.
This article was updated on August 13, 2026 to include information about the prior review by third-party experts and additional comments from an X spokesperson.
