Unveiling the Shadows: X Tests a New Tool to Clear Up ‘Shadowban’ Myths
X is stepping into the light with a new tool aimed at demystifying “shadowbans” and offering users insights into how their posts are treated by its algorithm. This experimental feature, currently being rolled out to a select group of testers, allows users to access a dedicated “under the hood” section in their app settings. Here, they can download and analyze data related to their account, shedding light on any potential visibility restrictions affecting their content.
Alongside this announcement, X revealed a significant shift towards openness, with an update that partially open-sources aspects of the ranking algorithm behind the widely used “For You” timeline. This move, described by Keith Coleman, X’s VP of Product, as “an unprecedented level of transparency into the X algorithm,” signifies a commitment to providing users with more clarity about their experiences on the platform.
However, utilizing this new feature comes with its complexities. Instead of delivering granular, post-by-post analytics, the tool offers aggregate data over the past month, detailing “visibility-impacting labels” connected to users’ accounts. As users sift through this information, it may not be as straightforward as anticipated. A recent example shared by popular user @cb_doge (aka Doge Designer) illustrated how the downloaded data is formatted in a .json file—appearing as a jumble of code that can be daunting to decipher. It includes important annotations, such as two posts marked with “NSFW” labels, restricting their visibility to non-followers and younger audiences.
This isn’t the first time X has attempted to provide transparency; prior releases of its recommendation algorithms yielded limited understanding of how the platform’s inner workings operate. However, this latest data dump begins to highlight the types of content deemed unsuitable for recommendations, featuring labels for various categories like spam, violence, impersonation, and hateful conduct. Interestingly, a unique label titled “for emergency use only” was also noted, referencing incidents requiring a specific response.
While the transparency efforts are commendable, it’s worth mentioning that there are no labels directly addressing political content. However, a “civic integrity” flag exists, which restricts the visibility of flagged posts to the author’s profile only. This is categorized for content reported to violate X’s Civic Integrity policies, indicating its reserved use for significant escalations.
In a bid to enhance clarity, X is contemplating allowing third-party recommender system experts to examine and provide feedback on this code release before it’s made widely available. Although the criteria for selecting these experts remains ambiguous, and it’s uncertain whether their evaluations will be public, significant plans for full transparency within X’s entire coding structure have been hinted at by Elon Musk, who mentioned a move toward open-source auditing in August.
Despite these steps towards openness, X admits it still withholds certain information that could potentially enable users to exploit the system, as well as the underlying codes governing advertisements and other elements beyond the “For You” timelines.
In summary, as X continues to evolve its platform, users are gaining a glimpse into the factors influencing their visibility. Whether this new tool will provide the clarity and understanding its users seek remains to be seen, but it undeniably marks a significant shift toward greater user empowerment and algorithmic transparency.