X has changed how its revenue-share programme treats stolen content. Where a large account reuploads a smaller account's post, the monetised impressions are now allocated entirely to the original creator — and the edits people used to defeat detection no longer work.
The numbers
- 1.5 million stolen posts detected in a single payout cycle.
- Over $1 million in payouts redirected to the creators of the stolen content.
- Payouts to aggregator accounts down roughly 80% over the year.
- The newest Grok model detects duplicated content at about three times the rate of its predecessor.
Adding watermarks, intros or other edits intended to pass reuploaded video off as original now sends the impressions to the original uploader rather than defeating the check. The same applies to copying viral text posts verbatim.
Why this is happening now
Content theft on X is not new, and the economics that drive it have been obvious since the revenue-share programme launched: paying accounts for impressions creates an industry in harvesting other people's work, because reuploading is far cheaper than making something.
The timing is explained by something else. X's feed is the primary data input for Grok and xAI's other work. A timeline saturated with duplicated, watermarked reuploads is degraded training data as well as a worse product. The company has an incentive to clean the corpus that is independent of fairness to creators.
That is worth stating plainly rather than cynically. The interests happen to align — creators want attribution, xAI wants clean data, users want a feed that is not the same clip eleven times — and an alignment of interests is a far more durable reason for a platform to act than goodwill. But it does explain why this arrived now, after years of complaints, and it is the kind of thing a reader should know when judging how permanent the change is.
The part that has not been answered
Automated attribution at this scale will produce false positives, and none of the reporting describes an appeals process.
The ambiguous cases are not rare. Two people filming the same public event produce near-identical footage. A reaction video legitimately contains the original. Commentary quoting a viral post is the normal function of a discussion platform. Licensed reuse looks exactly like theft to a duplicate detector. In each case the system must decide who gets paid, and a wrong call takes money from someone who earned it.
A detector running three times hotter than its predecessor will catch more theft and also flag more of these. Without a published appeal route, the accuracy of the classifier is the whole policy.
What it means if you post
Practically: there is now less reason to stay quiet about a reupload, because the mechanism redirects impressions rather than merely removing the copy. And the old defensive habit of stamping your own watermark across your work matters less for attribution than it did, since the detector reads through exactly that.
Sources: X's announcement; Social Media Today; TechCrunch; Digital Trends; Mezha; and Prothom Alo's original report.
