X previews labels that explain why its AI flagged a post

X is developing labels that would also say whether AI was used to make a post and how, Social Media Today reports, so creators can contest a flag with specifics. X hopes that feedback will help train its detector, Social Media Today reports; O’Herlihy said X leans on automated tagging because it is faster. Adweek’s example shows the need: the system once tagged a SpaceX launch photo as adult content, and no one, the poster included, could see the reason.

The pitch is commercial too. About half of X’s ad revenue has gone since Elon Musk’s 2022 takeover, by Adweek’s account, and brand safety is the pitch X is using to win it back. O’Herlihy pitched it as moderation that shows its workings, checked by outside auditors.

The tools that make AI ads leave their own marks. In our rights dataset, Google’s Veo always embeds SynthID, its invisible watermark, with the visible one a toggle, while our dataset rates Grok Imagine’s ad rights unclear, because a clause in its terms (archived copy, 21 September) can be read to require a “Created with Grok” credit.

X’s own example, a rocket photo flagged as adult content with no reason anyone could see, is the case these labels target: a brand whose post is flagged as AI-made would see what was detected and how far its reach was cut, and could appeal on those points.