The Messiness Premium

Polish used to prove effort. Now it indicates a machine make it.

The video was good. Technically beautiful: color graded, well paced, the audio clean, a soft push-in on the close that we’d argued about for twenty minutes and finally got right. We sent it over feeling good about it. The feedback came back in one line. “It feels kind of AI?”

Then the part that actually stung. Someone on the team had shot a thirty-second outtake on a phone, vertical, half blown out by a window behind the subject, with one stumble over a word that never got cut. They’d posted it as a throwaway. It outperformed the finished piece by a wide margin. The rough one tested better. The one we sweated over tested like a suspect.

If you make things for a living, you’ve had some version of this week. The instinct is to file it as a fluke, or as audiences getting lazy, or the algorithm doing something dumb. It’s none of those. What polish proves changed, and this video is still trying to prove the old thing.

Polished used to equal proof

For my whole career, high production value was the goal. A clean edit, a mood-appropriate color correction, a proper audio mix: those things cost money and time and skill, and audiences read them, correctly, as evidence that somebody had spent money and time and skill. “Polish” was a signal, and the signal said someone competent worked hard to do this right.

That signal has now flipped. Not weakened. Inverted. When a machine can produce a flawless edit for free in an afternoon, flawlessness stops being evidence of effort. It becomes evidence of nothing, or worse, evidence that nobody was there at all. The same gloss that used to say “professional” now whispers “generated.” Polish went from your credential to your alibi.

The audience already voted

The numbers are moving the same direction, and fast. Three years ago, audiences preferred AI-generated creator content at around 60%. By this year that preference had fallen to 26%. That’s not a soft drift. That’s a whole category being abandoned. And that shows up in actual money: when Aerie made a public pledge against AI-altered imagery, the post became its highest-performing of the month and drove something like $519K in earned media value. They got paid, in attention, for promising to leave the retouching off.

One study isolated the thing almost too cleanly. Researchers showed people identical pieces of art. The only difference was the label: human-made or AI-made. The work tagged human struck people as deeper and was judged worth more, by real margins. Same pixels. And the mechanism underneath it is the part that should change how you work: when people believed a human made it, higher perceived *effort* raised how much they liked it. When they believed AI made it, effort did nothing. The effort premium exists only for human work. Machines get no credit for trying, because we don’t believe they tried.

That’s the whole game in one finding. Effort is the scarce signal now. And messiness, the stumble, the window glare, the sentence that doubles back on itself, is the most legible evidence that a human actually spent themselves on the thing. The roughness isn’t the cost of being real. It’s the proof of it.

You can’t fake your way in

Which is exactly where this goes wrong, and I want to be careful, because the wrong lesson is an expensive one to learn.

The lesson is not “make it messy.” The second this becomes a tactic, it dies, and there’s solid research on why. Imperfection works as a signal of authenticity for one reason: people optimizing purely for impressions won’t deliberately choose the worse-looking option. The roughness is believable because a faker wouldn’t pick it. The moment faking it becomes the strategy, the moment every brand starts manufacturing its shaky handheld and its candid blooper reel, the signal collapses, because now the fakers are choosing it too.

And it’s already collapsing. The “handmade,” deliberately-imperfect look is now the mainstream brand answer to AI sameness, which means it’s saturating on the same compressed timeline everything else does. Gen Z, the audience most fluent in this stuff, is already rejecting performed authenticity as one more corporate costume. They can smell a manufactured stumble. We taught them how.

So you can’t fake your way into the messiness premium, because the premium is paid for the one thing faking removes: that a specific person was actually there, actually spent something, and left the evidence in the frame.

This is not an argument for sloppiness, and that distinction matters more than anything else here. The shaky video works because the person in it is real and present, not because shaky is good. Bad lighting on a stock-feeling testimonial is just bad. The premium isn’t on low quality. It’s on visible humanity, and those are not the same axis.

Where to leave the roughness

So the call isn’t “lower your production values.” It’s “figure out which of your production values were buying you credibility and which are now buying you suspicion.” Strip the polish off the places where polish reads as machine: the founder’s note, the customer story, the quick reaction to something that happened this morning. Leave the stumble in the smartphone video, because the stumble is the part a machine can’t honestly produce. And keep your competence where competence still signals care, in the work itself, the product, the thinking, the parts where getting it right *is* the human effort rather than something you’re hiding.

Go back to the last thing you made that “felt AI.” I’d bet the problem wasn’t that it was bad. The problem was that it was too clean... too sterile to prove a person had been in the room. That’s the catch worth making before the next thing ships: the roughness you keep sanding off, the stumble, the glare, the human evidence, is the exact thing you’ve been calling a flaw. It was the proof you were real, and you were editing it out.