Remember the “Cake or Not” trend? Someone hands you a perfectly ordinary object — a sneaker, a phone, a houseplant — and asks if it’s real or actually a hyper-realistic cake. The whole appeal was that your eyes, which you’d trust completely in any other situation, suddenly couldn’t be trusted at all.
We’re watching the exact same thing happen with video right now. Except instead of cakes, it’s “AI or Not” — and instead of a fun party trick, it’s becoming a real question every marketer, agency and viewer has to start asking.
Why this happened so fast?
A year or two ago, spotting AI video was almost too easy — a flickering hand, a face that morphed mid-sentence, background objects that didn’t quite make physical sense. Those tells were reliable enough that “is this AI” barely needed asking.
That gap has closed fast. Skin texture holds up under close inspection now. Micro-expressions land naturally. Product interactions — someone actually picking up and using an object — used to be the hardest thing for AI video to fake convincingly, and even that’s starting to hold together. Things in video that used to give it away are disappearing one by one, the same way “cake or not” got harder as bakers got better at replicating grain, texture, and imperfection.
The difference that actually matters?
Here’s where the analogy breaks, though, and it’s worth sitting with the difference: nobody’s harmed by a cake fooling them for ten seconds. It’s a party trick with zero stakes. “AI or Not” for video is a different question entirely, because video carries trust in a way a fake cake never had to — an ad implying a real customer using a real product, a testimonial that looks like an unscripted moment, a “behind the scenes” clip that never actually happened.
The guessing game is fun right up until the video is doing something other than entertaining — until it’s selling, persuading, or standing in for evidence.
What this means practically?
For anyone producing content with AI video (us included, building Fabeo), I think this cuts two ways:
It’s genuinely exciting from a craft standpoint. The tools have gotten good enough that “is this AI” is a real compliment about production quality, not a giveaway of a shortcut.
It raises the bar on disclosure and intent, not just fidelity. The more convincingly a video can pass as real, the more it matters why it was made and whether the audience knows what they’re looking at. A brand using AI video to test creative concepts fast is a very different use case from a video designed to make an audience believe something happened that didn’t.
What we’ve actually seen from customers?
This isn’t theoretical for us. A few patterns keep showing up in conversations with customers on Fabeo:
Some are genuinely proud their video is AI-generated and want that visible — for them, “made with AI, still looks this good” is the pitch, especially early-stage founders showing a product that doesn’t fully exist yet. Others want the option to remove any indication, usually for finished campaign work where they want the video judged purely on the story, not on how it was made. And a fair number sit somewhere in between — comfortable with AI doing the heavy lifting, but a little uneasy about a viewer not knowing at all.
That range is exactly why we default every video generated on Fabeo with metadata and a visible “AI-generated content” watermark — it stays on unless the customer explicitly chooses to remove it. The default is disclosure, not the other way around. It’s a small policy decision, but it’s really a bet on the point above: as the videos get harder to tell apart, trust becomes something you have to design for on purpose, not something you can assume survives on its own.
The closing thought
“Cake or not” stayed fun because the stakes never mattered. “AI or not” for video won’t get that luxury for long — and I think the businesses and creators who get ahead of that now, by being upfront about what’s AI-generated and what’s real, will end up trusted more, not less, as the guessing game gets harder for everyone.
Note: Drafted with help from Claude and Gemini — ideas, opinions, and the GPU bills are mine 😊