X is testing a label that explains why its AI limited your post

X is testing a label that explains why its AI limited your post

On Tuesday (Nov. 10), X showed labels that will appear under a post and explain what the automated moderation system detected, how much reach was limited and how to appeal. It's still a prototype, with no date and no list of countries. But for anyone who lives off clips and video on X, it's the first time the network has promised to say, post by post, why your numbers dropped.

Who presented it and what was on screen?

It was Michael O'Herlihy, a former Twitter employee who recently returned to X as head of safety, in a talk at the Brand & AI Safety Summit in New York. According to Adweek, the designs sit under the post and say what X's AI detected, how much reach was reduced as a result and how to appeal.

O'Herlihy called the idea a "glass box," as opposed to the black box that moderation usually is on social networks. The example he used is a good one because it's embarrassing: the system once flagged a photo of a SpaceX rocket launch as "adult content," and nobody could see why.

A caveat on how to read this. Social Media Today described the test as labels that say whether AI was used to create the post. Coverage from people who watched the talk says something different: the AI in question is X's moderation AI, and the label explains its decision. Both agree on the essentials: more detail on why the post was flagged, and room for the creator to appeal.

Why is X doing this now?

Because X has come to rely more on automated flagging. O'Herlihy said the platform increased its use of automatic tags for violations because it's much faster, according to Social Media Today. More automation means more errors at scale, and errors without an explanation turn into shadowban complaints.

There's a commercial side too. X's ad revenue has fallen by roughly half since Elon Musk bought the company in 2022, Adweek notes, and the talk was aimed at an audience of brands. Showing how moderation works from the inside is a sales pitch to advertisers who are afraid of appearing next to problematic posts.

What already exists today: Under the Hood

An explanation on the post itself is new, but X already shows part of this. Since August 2026, the Under the Hood tool, in pilot, lets anyone who posted at least 10 times in the month download a file with the labels that may have reduced the visibility of their account and posts in the previous month. In September, it also started showing when a limit came from a government's legal demand, along with the country that made it, as TechCrunch reported.

The problem with Under the Hood is the format: a technical, monthly file that not everyone gets. The new label puts the information where the creator is already looking, on the post, in the moment.

What changes for you

If you post rarely, almost nothing. For clipping channels, game highlight accounts, livestreams and podcasts, quite a lot, because these are the accounts where automated moderation is most likely to get it wrong: a clip out of context, background music, violent game footage, a caption with a word that trips a filter.

Three practical effects, if the design ships as shown:

• You'll know which post got hit. Today a drop in reach arrives as a mystery, and the usual reaction is to change everything at once. With the reason written out, you can fix only what was flagged.
• The label is public. Social Media Today points out that people who see a flagged post tend to share it less. In other words, the label isn't just a notice for the author: it weighs on distribution itself.
• Appealing starts to pay off. X expects to use appeals to train the detector. A false positive nobody appeals keeps teaching the machine the wrong thing.

This fits what we're already seeing on other networks: YouTube announced it can apply the branded content label on its own, as the blog covered in the post on undisclosed sponsored posts, and Instagram started weighing content provenance signals. The trend is the same: the network decides on its own and, at best, explains afterward.

Does it apply in Brazil yet?

There's no confirmation yet. X only showed designs at an event in the United States, with no date, and didn't say in which countries the labels will appear. Under the Hood, which already works, is a pilot with limited access.

What to do now

I'd treat the announcement as a hygiene reminder, not a reason to change your strategy:

1. Check whether you have Under the Hood. If your account posted 10 times or more last month and made it into the pilot, download the file and match the labels against the posts that did worse.
2. Appeal anything flagged by mistake. A livestream clip with context, a game screenshot, a news image: if the system got it wrong, appeal. It will matter even more once the reason shows up on the post.
3. Keep the source of your material. If you clip other people's content, having the creator's permission or the link to the clipping program on hand makes an appeal easier. X, by the way, started paying both the clipper and the creator, and a clipping account flagged by mistake loses money on both sides.
4. Don't hide AI where it shows up. If the video has a generated face, voice or scene, say so clearly in the post text. Networks are cracking down on fully generated video without disclosure, and faceless channels with a persona are the first target.

If you clip livestreams or podcasts for several networks, what protects your account most is making sure the clip makes sense on its own, without depending on context the machine can't see. In Cut.Pro, the AI picks the segments from the long video and you review each clip and its captions before publishing, which helps you catch the ambiguous segment before X's detector does.

When the label actually ships, I'll come back with the exact text and where it appears in the app. Until then, what exists is Under the Hood.