Artificial intelligence generated images, video and audio have become so realistic that it is often difficult for social media users to tell what is real and what is fake. More than a dozen online detection tools claim they can identify artificial intelligence content by scanning for hidden watermarks and digital irregularities. Tests conducted by The New York Times found that while many tools can flag obvious fakes, none are consistently reliable enough to deliver definitive judgments.

The detectors performed best on simple artificial intelligence generated images and on clearly authentic photos and videos. They were less effective when analyzing complex or highly realistic fabricated scenes, and only a handful could meaningfully assess video content. Audio detection showed stronger results, with several tools accurately identifying synthetic voices even when clips were modified.

A major weakness emerged when real images were subtly altered with artificial intelligence elements, as most tools failed to catch these edits. Experts say detection systems are locked in a continuing technological race with increasingly sophisticated generators, meaning users and fact checkers should treat results as one signal among many rather than final proof.