Zero. That is the total amount of creative merit found in a fully AI-generated “satisfying” video. We’ve all seen them—those uncanny, glossy clips of people making “coffee” out of soap or building impossible houses out of mud, narrated by a voice that sounds like a bored GPS. It is digital noise, pure and simple. It is content designed not for humans, but for the specific vulnerabilities of an algorithm that rewards watch-time over actual value.
Snapchat has finally decided to stop fueling the fire. According to TechCrunch, the company is tweaking its recommendation systems to ensure that fully AI-generated content is no longer eligible for Spotlight recommendations. They aren’t banning the stuff—you can still upload your weird Sora-esque fever dreams if you really want to—but they’ve stopped giving them the algorithmic megaphone.
It is a blunt move. For the last two years, the incentive structure of short-form video has been a race to the bottom. If you could prompt a video into existence in ten seconds that looked “satisfying” enough to keep a teenager scrolling for five more seconds, you won. The result was a flood of content that felt like it was written by a committee of bots trying to guess what humans like, without ever having been a human.
It’s about time.
The real question isn’t whether this is a good move—it obviously is—but whether it’s even possible to enforce. We are entering the era of the “AI-assisted” gray area, and this is where things get messy. There is a massive difference between a video generated entirely from a prompt and a video shot by a human that uses AI for color grading, noise reduction, or maybe a subtle background swap to hide a messy bedroom.
Who decides where “human-created” ends and “AI-generated” begins? If a creator uses an AI tool to script the video but films it themselves, does that count? If they use a deepfake of their own face to fix a blink or a stray hair, are they suddenly “slop”?
Trying to filter for “humanity” in a stream of pixels is like trying to tell the difference between a hand-formed burger and a frozen patty that’s been carefully seared to look artisanal. You might get it right most of the time, but eventually, the industrial version gets so good that the taste test fails.
The friction here is technical. Detection models are notoriously brittle and expensive to run at scale. Every single upload has to be passed through a classifier (which costs GPU cycles and adds latency), and those classifiers are playing a game of whack-a-mole where the moles have better hardware than the players. Snapchat is betting that their internal classifiers can keep up, but we’ve seen how these things go. The “AI-generated” label is usually a lagging indicator, not a real-time shield.
Still, the strategic pivot is correct. The “dead internet theory” isn’t a conspiracy anymore; it’s a user experience problem. If a platform becomes 90% synthetic, the humans leave. And once the humans leave, the advertisers leave, because nobody pays to show ads to other bots. It is a death spiral of synthetic engagement.
By Q4, we will see TikTok and Instagram forced to implement similar “human-only” visibility boosts. They can’t afford to let their feeds turn into a mirror room of synthetic hallucinations. If they don’t, the platforms will simply become archives of AI-generated landfill.
(I might be overestimating how much users actually care, but the churn is real).
Does anyone actually enjoy watching a synthetic person describe a fake product in a fake room?
Probably not.
