X updates “Under the Hood” transparency report, allowing users to view flagged posts
X announced on October 30 through its open-source account that it had updated the “Under the Hood” transparency report, and eligible users can now see which specific posts were automatically flagged by the system as potentially containing spam or other visibility-restriction issues.
Eligible accounts must have been registered for at least one year and must have posted 10 or more posts last month. The report is generated on a best-effort basis for the previous calendar month, and the related build code is open source. When a label affects less than 5% of posts in that month, the system lists the specific post IDs so users can view them directly. Common labels include SPAM_HIGH_RECALL, which means the post was automatically detected by the system as potentially spam, and the effect is that recommendations are hidden from non-followers (followers can still see it in Following). A previous update had already included information on content restricted for government-request reasons.
Creators have begun sharing their personal report results. Some accounts show very few or no labels, such as 0.16% or completely clean, while others found individual posts flagged. The feature responds to users’ demand for transparency around visibility restrictions and helps explain the impact of platform decisions on reach to non-followers. X did not claim that the report covers real-time or all-account status, and it also did not provide detailed updates on an appeal mechanism.
Original sources:
Title: Under the Hood now shows you which posts received a visibility-limiting label... Publisher: X Open Source (@XOpenSource) Publication date: 2026-10-30 URL: https://x.com/XOpenSource/status/2116262141745565826 Source nature: official platform account post Verification status: full post and subsequent user shares were read in this review, verification date 2026-10-31 Supporting content: confirmed that the update can show the specific flagged posts (listing them when they account for less than 5%), and thanked users for feedback.
Title: X Under the Hood Report: How to Get It and Read It Publisher: ClimbX Publication date: 2026-10-17 URL: https://climbx.so/blog/x-under-the-hood Source nature: third-party analysis (based on platform explanations and open-source code) Verification status: the report explanation and examples were read in this review, verification date 2026-10-31 Supporting content: confirmed the eligibility conditions (account older than one year, more than 10 posts last month), the meaning of SPAM_HIGH_RECALL and its effect on recommendations to non-followers, and that the report covers the previous month. It does not prove specific user proportions or the effects of all labels.
Title: X's Under the Hood (2026): The Official Report, Explained Publisher: Shadowban Radar Publication date: 2026-10-02 URL: https://shadowbanradar.com/x-under-the-hood Source nature: third-party explanation Verification status: the feature timeline and content description were read in this review, verification date 2026-10-31 Supporting content: confirmed that the report is a transparency tool visible only to account owners for the previous month and that it has included government-required content filtering information since September.
ABAB AI Insight
X launched the Under the Hood report in August 2026, expanded it to government-required content filtering in September, and provided a more accessible page on September 24. This update further responds to creators’ feedback that they are being demoted without knowing which specific post is affected, and it lists specific post IDs when the affected share of a label is below 5%. Related algorithms and label logic are partially open source, and users can view their personal reports through a direct link.
In terms of functionality, the report aggregates posts and account-level labels by calendar month and explains the impact on visibility, such as hiding recommendations from non-followers. SPAM_HIGH_RECALL is a high-recall automatic detection label that may cover engagement bait, rapid replies, or low-quality links, but it does not mean the account is necessarily restricted or completely hidden. Government-related restrictions are separately marked with the country of origin and legal requirement. X emphasizes that the report is best-effort and does not constitute real-time monitoring or a complete audit log.
A historically comparable case is other platforms providing creators with limited explanations of why a piece of content was demoted, but most only give broad reasons. X differs in that it binds labels to specific post IDs, within the threshold, and combines that with open-source code, allowing creators to compare it against their own content. A similarity is that the automated system still dominates labeling, and users cannot directly change historical labels.
This update changes the observability of creators’ own visibility restrictions: from only knowing the proportion to being able to locate specific posts. For high-frequency creators, low proportions of labels may be treated as noise; for accounts that rely on recommendation-based reach, it provides a basis for adjusting engagement or linking strategies. Observable metrics are the distribution of label proportions and types in subsequent user shares, and whether X further opens appeals or real-time queries.
ABAB News · Cognitive Law
Showing specific posts does not mean the label has been proven accurate.
Hiding recommendations from non-followers does not mean followers cannot see them.
The report covers the previous month and does not reflect the current real-time status.