In the past, fake videos were easily noticeable. Recently? I'm much worse at identifying them. Thanks to the fine folks at NPR, we have the quiz results to prove it. (ah.)
A short four-question test asks you to distinguish between AI and human-generated videos. Scenarios range from serious to unexpected to cute. All kinds of funny and surprising clips that go viral in group chats and social media. Given the improvements in AI-generated videos, I expected there to be at least one mistake. I have failed many more times. (But thankfully not all four.)
I doubt most people would do better. At least not without coaching.
Sure, the AI produces a lot of slop. But the accuracy of video generation has also improved, with far fewer immediately obvious errors. If you watch a video with the default expectation that most of it is real and fakes are easy to spot, you might stumble like I did. I went into the quiz with the assumption that the AI video would be quickly lost due to a technical error.
Instead, you should pay attention to more subtle cues, at least relatively speaking. Details matter. Context matters. Experience and expertise can help you determine what is reasonable and what is not. In other words, identifying AI videos uses similar skills as identifying fraud.
A simple watermark can make a big difference in helping people avoid AI slops. Or at least help you understand the nature of what you're reading, listening to, or watching.
digiarty
Too good to be true? Please proceed with caution. Are you trying to exploit your emotions for better or worse? The point may be to depict a strong reaction that overturns rational thought. And what if someone asks for money? Be sure to stop and check the legitimacy of the video.
The NPR quiz gives you more specific tips on how to recognize AI video, including clip length, framing, and lighting. Our guide on how to spot fake AI videos includes even more detailed advice, covering elements such as the very basic (but easily skipped) steps of checking physics, soundtracks, and even metadata.
Video used to be the most reliable documentation format. Given the slowing rate at which platforms are being taken over, AI threatens that trust. Until more laws are passed to help identify AI-generated materials, we all need to keep a sharp eye.
California’s AI Transparency Act, which was originally scheduled to require watermarks and other identifiers on AI-generated or altered text, images, audio, and video starting January 1, 2026, has been delayed until August 2, 2026.
There are many months left to avoid the AI slope on your own.
