X users will soon have a way to save their complaints about “shadowbans” affecting their reach on the service. The platform is testing a new tool that gives users details about how their posts are processed by the algorithm, including whether they have been limited in any way.
The feature, which is currently only available to a “small group” of testers, is available through a new “under the hood” page in the app settings. People who have access to the tool can download the data associated with their account and browse it to find “visibility-limiting tags” information affecting their posts.
X announced this feature alongside an update that opens up more aspects of the ranking algorithm that powers the “for you” timeline. Keith Coleman, vice president of product at X, called the changes “an unprecedented level of transparency in the X algorithm.”
But in reality, using this feature can be a little more complicated than expected. For starters, downloading data via “under the hood” does not provide statistics for each post. Instead, it “lets people see aggregate statistics” from the past month on “which tags impact visibility on their account and posts,” according to details shared by X on GitHub.
And the information it provides may be difficult for some to parse. An example, shared by @cb_doge (also known as Doge Designer), of what the data looks like “under the hood”. In the video, the download results in a .json file which, at first glance, may look like a bunch of code. Look closely and you can see that two of the account’s posts were hit with two separate “NSFW” tags, resulting in those posts being hidden from non-followers and underage users.
Screenshot via X
X already had open source versions of its recommendation algorithm, but these code dumps revealed very little about how the platform actually worked. But the latest version provides insight into the type of content that X deems ineligible for recommendations. For example, the GitHub repository lists many tags related to different types of NSFW posts as well as those relating to spam, violence, impersonation, and hateful behavior. There are also more specialized labels, such as one called “for emergency use only”, which is described as an “emergency incident response label with personalized notice”.
There does not appear to be a label related to political content, although there is a “civic integrity” indicator that restricts the visibility of a post to its author’s profile only. The label is used for posts “detected in response to a user report containing content in violation of X’s Civic Integrity Policy.” According to company policy, this label is “reserved for critical escalations.”
In addition to “under the hood,” X said “third-party recommendation system experts” were able to preview and test this code version in advance.
The company’s latest version doesn’t accomplish exactly that. In fact, he notes that it continues to withhold certain information that could potentially be used to “cheat the system” and manipulate the algorithm. X also hasn’t released the underlying code that deals with ads and other systems outside of its “to you” timelines. An X spokesperson explained that future updates could include additional labels related to brand safety.
Updated, August 13, 2026, 4:40 p.m. PT: This post has been updated to clarify that third-party experts have already reviewed X’s version and to add additional information from an X spokesperson.
