X to explain AI moderation: new labels are meant to show reasons for restricted posts

X wants to make its AI moderation easier to understand. At the Brand & AI Safety Summit in New York, on 10 2026 November, the platform presented a new direction for its moderation labels: users should be able to better understand why automated systems flag a post. For creators and brands, the key question is this: what happens to visibility when the algorithm classifies a post as problematic?

The move was picked up by ADWEEK. X presented the topic as part of the talk "Show Your Work: Making Transparency the New Standard for Safety" by its new head of safety, Michael O’Herlihy. The official conference agenda lists him alongside Mike Zaneis of the Trustworthy Accountability Group. This is a preview, not the announcement of a completed worldwide rollout.

From moderation decision to a traceable explanation

X has already been running "Under the Hood" since August 2026, a transparency area for labels that can affect how accounts and posts are distributed. The system makes visible restrictions that would otherwise only show up as, for example, unusually low impressions. In September, notices about blocks and reach restrictions due to local laws were added. We explained the background to this expansion in our article on X and legally limited reach.

Now the reason for an automatic label takes centre stage. For social teams, this is far more helpful than a bare error message: whether a post is treated as possible spam, downranked because of sensitive content, or restricted under a policy determines how you review the case and whether you appeal.

In its own Brand Safety Help Center, X describes that moderation and ad environment controls are built from machine learning, human review and context signals drawn from text, images and metadata.

Why advertisers care about the labels

For ad sales, explainability is part of the safety strategy. According to its own documentation, X already offers exclusion lists with up to 4.000 keywords and 2.000 accounts, as well as two sensitivity levels for the immediate ad environment: "Standard" and "Limited". The platform also works with Integral Ad Science and DoubleVerify to verify ad environments.

This creates two different types of control: publishers and creators check which labels affect their own content. Advertisers control which content their ads appear next to. The interplay becomes important for every major campaign, especially in news and debate environments. Our overview of the TikTok Brand Safety Center shows how other platforms structure the topic.

These developments fall into a period of intense regulatory scrutiny. Since January, the European Commission2026 has been examining, under the Digital Services Act, risks including those linked to Grok features and X's recommender systems. Transparency about moderation decisions is therefore a relevant topic well beyond ad sales.

What German social teams should check now

If X is an important channel for you, you can already compare the available Under the Hood reports with your regular analytics today. Posts with noticeably lower reach among non-followers and posts whose visibility suddenly drops are especially revealing. In such cases, document the affected post, the labels shown and the time period; if you spot an apparent misclassification, the official appeal route is worth using.

On the advertiser side, a check of the sensitivity settings and adjacency controls belongs in the campaign setup. Reporting should clearly separate platform-internal safety systems from external verification. The conflict between safety rules and demands for platform data is also not new, as the Ofcom dispute with Meta, TikTok and X shows.

The explanation feature now previewed will hinge on how specifically X discloses the causes of its AI moderation decisions and when the labels become available to users. Until then, the announcement is mainly a signal: transparent moderation is increasingly becoming a product feature for social platforms, not just an internal safety task.

Sources

• ADWEEK, 10 November 2026: X Previews Labels That Tell Users Why Its AI System Flagged Their Posts
• Brand & AI Safety Summit New York: Programme of 10 November 2026
• X: Products That Protect – Moderation, Under the Hood and Ad Controls
• X: Notices on X and what they mean
• European Commission: DSA investigation into Grok and X's recommender systems