X Will Show Users Why Its AI Flagged Their Posts

X previewed labels on Tuesday (Nov. 10) that would tell users exactly why its AI flagged their posts. The move comes after the system labeled a SpaceX rocket launch image as "adult content" — and nobody, not even the post's author, could find out why.

Michael O'Herlihy, X's new head of safety, showed the designs at the Brand & AI Safety Summit in New York. He is a Twitter veteran who recently returned to the company. The labels would sit beneath flagged content and do three things, according to the preview:

• Explain what the AI detected in a post
• Detail how the post's reach was limited after the flag
• Let the user appeal the decision

Why does this matter to marketers?

Because X needs advertisers back. The company's ad revenue has plummeted since Elon Musk took over, and brand safety is now the core pitch to win brands over. Content moderation that advertisers can't see or verify is a hard sell; moderation that explains itself, in theory, is an easier one.

O'Herlihy's argument goes further. He said AI-led moderation creates auditable records — permanent, inspectable logs of each decision — better than human review does. Human reviewers leave notes; an AI pipeline leaves a traceable decision path, or at least that's the claim.

Here's the caveat. An auditable record is only worth what the underlying decision is worth. The SpaceX rocket incident proves the point: X's system made a wrong call, and the audit trail, as it exists today, told the user nothing. Transparency labels fix the second problem. They don't fix the first.

What did the preview actually show?

Designs, not a shipped product. O'Herlihy "suggested" the systems "may" show their decisioning process in the future, per his on-stage remarks. There is no launch date. There is no detail on which classification categories the labels would cover, how quickly appeals would be resolved, or whether human reviewers sit anywhere in the loop.

What's not said matters as much as what is. X did not commit to publishing error rates for its AI moderation. It did not say whether flagged users get reach restored after a successful appeal. And it did not address how labels apply to advertisers' own promoted posts — arguably the case brands care about most.

For comparison, think of a credit card decline. Banks must tell you a reason code and give you a dispute path. X is proposing roughly the same standard for algorithmic content decisions: show the reason, show the penalty, open an appeal.

Is this a pattern or a one-off?

The timing fits a broader playbook. Platforms under ad-revenue pressure tend to rediscover transparency when brands walk. X's revenue collapse gives it more incentive than most to prove moderation is accountable and brand-adjacent content is protected.

But a preview at an industry summit is a promise, not a product. CMOs evaluating X as a channel should treat the labels as a signal of direction — and wait for a ship date, published accuracy numbers, and a real appeals track record before pricing that transparency into media plans. The rocket photo got flagged by mistake. The next question is whether the label explaining that mistake arrives before or after the next advertiser budget does.