Users on X have long suspected their posts vanish from feeds without explanation. Now the platform offers a way to check. A fresh pilot feature called Under the Hood lets select accounts download data on visibility-limiting labels applied to their content over the past month.
The rollout comes as X open-sources additional code for its For You timeline ranking. Keith Coleman, X’s VP of product, described the moves as “an unprecedented level of transparency into the X algorithm.” Engadget reported the development on August 13, 2026, just as the company shared details on GitHub.
Access remains limited. Only a small randomized group of accounts at least one year old and active with at least 10 posts in the prior month can see the new settings page. Those users download a JSON file packed with aggregate statistics rather than post-by-post breakdowns. The file reveals how many times certain labels triggered restrictions. One example shared by influencer account @cb_doge showed two posts flagged with NSFW labels. Those posts stayed hidden from non-followers and users under 18.
But. The data demands technical comfort. Raw JSON output reads like code at first glance. Casual users may struggle to interpret counts of labels tied to spam, violence, impersonation or hateful conduct. X lists dozens of such flags in the newly released code. A specialized “for emergency use only” marker triggers custom notices during major incidents. Another, civic integrity, confines a post to the author’s profile alone when it violates election-related rules.
X insists it does not shadowban. Replies from its official AI account Grok on August 14, 2026, repeated the company line. “X does not shadowban. Official policy remains: ‘Simply put, we don’t shadow ban! Ever.’” Ranking for relevance and quality can push replies under “Show more replies” or reduce notifications. Followers still see posts on the profile itself. The distinction matters. Many creators treat any drop in reach as de facto censorship.
Third-party checkers already flood the market. Sites such as Sorsa’s free shadowban tester and Circleboom’s tool query public search and suggestion endpoints to detect search bans or reduced visibility. Guides published in 2026 from Multilogin and others advise combining these tests with incognito searches. Yet none match the authority of data pulled straight from X’s systems.
The pilot builds on years of complaints. During the pre-Musk Twitter era and after, political accounts, marketers and everyday users reported sudden engagement collapses. TikTok creators voiced parallel frustrations in a June 2024 FWIW newsletter, citing anecdotal shadowbans on words like “vote” or “abortion.” X’s new labels avoid overt political categories. No dedicated flag for partisan speech appears in the open-sourced list. The civic integrity label applies only to critical escalations, according to the company’s election integrity policy.
Transparency has limits. X withholds certain ranking signals that could let bad actors game the system. The released code covers For You recommendations but skips ads and other surfaces. A company spokesperson told Engadget that future updates may add brand-safety labels. Elon Musk posted in early August 2026 that every line of code touching X systems would become open source and third-party audited. The current drop falls short of that promise. Researchers previously found earlier algorithm releases revealed little about actual operations.
And the timing feels strategic. X faces persistent questions over content moderation after Musk’s acquisition. By handing users their own label data, the company shifts some burden of proof. Creators can now point to concrete NSFW or spam flags instead of vague impressions of suppression. Whether the tool quiets complaints remains uncertain. Aggregate monthly numbers lack the granularity many demand. A single controversial post might carry a label that tanks distribution for days, yet the report only tallies totals.
Independent developers already dissect the GitHub repository. The under-the-hood labels file details exactly how visibility changes. NSFW variants hide content from broad audiences. Spam labels suppress repeat offenders. Violence and hateful conduct markers trigger similar curbs. These mechanics echo what moderation experts have described for years. Platforms apply thousands of such signals daily. Most users never learn which ones hit them.
X’s approach differs from outright bans. Affected posts don’t disappear. They simply reach fewer eyes. Followers see them. Search visibility or recommendation to strangers drops. The effect mimics a shadowban even if the company rejects the term. Grok’s responses on X yesterday stressed that ranking differs from bans. “Ranking for relevance is not a shadowban.” Quality filters in notification settings can further mute reach without any formal penalty.
Early tester feedback trickled out on the platform itself. Some posted screenshots of their JSON files showing zero labels. Others discovered multiple flags they never suspected. The pilot could expand quickly if uptake proves positive. X has not detailed criteria beyond account age and activity. Nor has it said when or whether the feature reaches all users.
Broader implications stretch beyond individual complaints. Advertisers care about brand safety. Regulators watch algorithmic accountability. European rules and U.S. debates over Section 230 keep pressure high. By open-sourcing parts of the model and exposing labels, X signals cooperation. Skeptics note the gaps. Code for ads stays closed. Some ranking logic stays proprietary. The company still decides which labels apply and when.
One clear winner emerges. Power users and researchers gain data to test hypotheses. Does posting at certain hours trigger filters? Do specific phrases invite NSFW review? Aggregate stats won’t answer every question. They do provide a baseline absent until now. Combined with third-party tools that check search suggestion bans, users assemble a fuller picture.
Critics argue any visibility limit equals censorship. Supporters counter that unchecked spam and abuse would render the platform unusable. The Under the Hood tool won’t settle that debate. It does give participants hard numbers instead of anecdotes. For an industry built on attention, those numbers carry weight.
Expect iteration. X already plans to refine the pilot based on tester input. Additional label types may appear. Per-post visibility reports could follow if demand grows. For now the feature marks a modest but tangible step. Users tired of guessing why their reach collapsed finally hold a report in their hands. Even if reading it requires squinting at JSON.