Fooled by fake videos? Unsure what to trust? Here’s how to to tell what’s real.

There’s a term for artificially generated content that permeates online spaces — creators call it AI slop, and when generative AI first emerged back in late 2022, that was true. AI photos and videos used to be painfully, obviously fake. The lighting was off, the physics were unrealistic, people had too many fingers or limbs or odd body proportions, and textures appeared fuzzy or glossy, even in places where it didn’t make sense. They just didn’t look real.

Many of you probably remember the nightmare fuel that was the early video of Will Smith eating spaghetti. It’s terrifying.

This isn’t the case any more. In just two short years, AI videos have become convincingly realistic to the point that deepfakes — content that perfectly mimics real people, places, and events — are now running rampant. For just one quick example of how far AI videos have come, check out Will Smith eating spaghetti, then and now.

None of it is real unless it is verifiable, and that is becoming increasingly hard to do.

Even the Trump administration recently rallied around AI-generated content, using it as a political tool to poke fun at the left and its policies. The latest entry portrayed AI Hakeem Jeffries wearing a sombrero while standing beside a miffed Chuck Schumer who is speaking a little more honestly than usual, a telltale sign that the video is fake.

While some AI-generated videos on the internet are simple memes posted in good fun, there is a darker side to AI content that makes the internet an increasingly unreliable place for truth, facts, and reality.

How to tell if an online video is fake

AI videos in 2025 are more convincing than ever. Not only do most AI video platforms pass the spaghetti-eating Turing test, but they have also solved many of the issues that used to run rampant (too many fingers, weird physics, etc.). The good news is that there are still a few ways to tell an AI video from a real one.

At least for now.

First, most videos created with OpenAI Sora, Grok Imagine, and Gemini Veo have clear watermarks stamped directly on the content. I emphasize “most,” because last month, violent Sora-generated videos cropped up online that didn’t have a watermark, suggesting that either the marks were manually removed or there’s a bug in Sora’s platform.

Your second-best defense against AI-generated content is your gut. We’re still early enough in the AI video race that many of them still look “off.” They have a strange filter-like sheen to them that’s reminiscent of watching content in a dream. Natural facial expressions and voice inflections continue to be a problem. AI videos also still have trouble with tedious or more complex physics (especially fluid motions) and special effects (explosions, crashing waves, etc.).

RELATED: Here’s how to get the most annoying new update off of your iPhone

Photo by: Nano Calvo/VW Pics/Universal Images Group via Getty Images

At the same time, other videos, like this clip of Neil deGrasse Tyson, are shockingly realistic. Even the finer details are nearly perfect, from the background in Tyson’s office to his mannerisms and speech patterns — all of it feels authentic.

Now watch the video again. Look closely at what happens after Tyson reveals the truth. It’s clear that the first half of the video is fake, but it’s harder to tell if the second half is actually real. A notable red flag is the way the video floats on top of his phone as he pulls it away from the camera. That could just be a simple editing trick, or it could be a sign that the entire thing is a deepfake. The problem is that there’s no way to know for sure.

Why deepfakes are so dangerous

Deepfakes pose a real problem to society, and no one is ready for the aftermath. According to Statista, U.S. adults spend more than 60% of their daily screen time watching video content. If the content they consume isn’t real, this can greatly impact their perception of real-world events, warp their expectations around life, love, and happiness, facilitate political deception, chip away at mental health, and more.

Truth only exists if the content we see is real. False fabrications can easily twist facts, spread lies, and sow doubt, all of which will destabilize social media, discredit the internet at large, and upend society overall.

Deepfakes, however, are real, at least in the sense that they exist. Even worse, they are becoming more prevalent, and they are outright dangerous. They are a threat because they are extremely convincing and almost impossible to discern from reality. Not only can a deepfake be used to show a prominent figure (politicians, celebrities, etc.) doing or saying bad things that didn’t actually happen, but deepfakes can also be used as an excuse to cover up something a person actually did on film. The damage goes both ways, obfuscating the truth, ruining reputations, and cultivating chaos.

Soon, videos like the Neil deGrasse Tyson clip will become the norm, and the consequences will be utterly dire. You’ll see presidents declare war on other countries without uttering a real word. Foreign nations will drop bombs on their opponents without firing a shot, and terrorists will commit atrocities on innocent people that don’t exist. All of it is coming, and even though none of it will be real, we won’t be able to tell the difference between truth and lies. The internet — possibly even the world — will descend into turmoil.

Don’t believe everything you see online

Okay, so the internet has never been a bastion of truth. Since the dawn of dial-up, different forms of deception have crept throughout, bending facts or outright distorting the truth wholesale. This time, it’s a little different. Generative AI doesn’t just twist narratives to align with an agenda. It outright creates them, mimicking real life so convincingly that we’re compelled to believe what we see.

From here on out, it’s safe to assume that nothing on the internet is real — not politicians spewing nonsense, not war propaganda from some far-flung country, not even the adorable animal videos on your Facebook feed (sorry, Grandma!). None of it is real unless it is verifiable, and that is becoming increasingly hard to do in the age of generative AI. The open internet we knew is dead. The only thing you can trust today is what you see in person with your own eyes and the stories published by trusted sources online. Take everything else with a heaping handful of salt.

This is why reputable news outlets will be even more important in the AI future. If anyone can be trusted to publish real, authentic, truthful content, it should be our media. As for who in the press is telling the truth, Glenn Beck’s “liar, liar” test is a good place to start.