There was a time when you didn’t have to tell anyone that a person made something.
That was simply the default.
The photograph was taken by someone. The song was recorded by someone. The article was written by someone. The person answering your email was, presumably, a person.
Human authorship wasn’t a selling point because it wasn’t a differentiator.
That’s the past. Today, the question isn’t simply whether something is good, useful, entertaining, beautiful, or persuasive. It’s who, or what, made it. And increasingly, we want proof.
I wrote about some of this last year in The New AI: Authentic Individual, where I argued that authenticity becomes more valuable when everything around us can be simulated. At the time, the interesting question was how we keep our humanity visible while using machines that can increasingly imitate it.
I think that question has a sharper follow-up now. The answer was never going to be as straightforward as choosing humans over machines, or machines over humans. The answer is making the human contribution visible, on purpose, in a way other people can actually check.
That’s a question about labels – so what happens when being human becomes something you have to label?
When Production Becomes Abundant, Provenance Becomes Scarce
We’ve been here before.
Organic became a label because industrial food production made the distinction matter. Local became a label because supply chains made distance invisible. Handmade became a label because machines made mass production easier. Fair trade became a label because consumers wanted to know something about the conditions behind the product, not just the product itself.
The label exists because something that used to be obvious no longer is.
AI introduced the same problem for human creatorship. A photograph can be generated without a camera. A song can be generated without musicians. An article can be generated without a writer. The output still exists. But the story behind the output has changed.
When anyone can produce almost anything, knowing who actually made it, and how, starts to matter more. The process becomes part of the story.
The Labels Are Already Here
This isn’t a thought experiment. It’s already happening.
British novelist Sarah Hall put a “HUMAN WRITTEN” maker’s mark on the cover of her novel, Helm, joining a small but growing wave of authors using human-authorship marks as AI-generated content becomes more common.
In April 2026, a verification initiative called The Human Made Mark launched for film and television, requiring producers to submit call sheets, behind-the-scenes photos, credit lists, and a signed legal declaration before attaching the mark to a production. Co-founder Eric Gruber described the ambition plainly: “The Human Made Mark is the Michelin Star of human craft.”
Fair Trade. Organic. Michelin Star. We already know how to build trust infrastructure around something that’s hard to see. We’ve just never had to build it around whether a human was in the room.
What We Already Know
A 2026 study in Frontiers in Psychology, based on a controlled experiment with 618 short-form-video users, tested how a “human-made” label, an “AI-generated” label, and no label at all affected how much effort viewers assumed went into a piece of content.
The AI-generated label significantly reduced perceived effort. The human-made label did not significantly increase it relative to unlabeled content.
The researchers read that as evidence of an implicit default: audiences already assume a human was behind the work unless told otherwise.
A separate 2025 study from researchers at NYU Stern and Emory found the same results from the other direction: disclosing that an ad was made with generative AI reduced advertising effectiveness by up to 31.5% in their experiments.
The research combined laboratory and field studies of visual advertising, so the finding shouldn’t be generalized to every form of AI disclosure. But it does provide evidence that the disclosure itself can change how an audience responds.
Today, human-made is mostly invisible because it’s assumed. The certification proponents above are betting that in some categories, that assumption is about to run out. Nobody’s proven the bet pays off yet. It’s a coherent bet, not a wishful one.
The infrastructure being built for this is already smarter than a binary AI/human stamp. C2PA and Adobe’s Content Credentials attach a record to a file describing what happened to it: when it was created, what tools touched it, what was edited afterward. That’s not “this was AI-detected.” It’s “here is what we know about this asset’s history,” which can describe human involvement, machine involvement, and the history of changes instead of forcing everything into one of two boxes. The IAB’s own framework works the same way, requiring disclosure only when AI use “materially affects authenticity, identity, or representation in ways that could mislead consumers,” not every time a tool touches something.
The more useful question is where human judgment enters the process. No one has agreed on a definition of “human-made” yet, and the label will probably keep evolving. The distinction that matters is whether a human makes meaningful decisions about the work, or simply accepts what the machine gives them.
But What Does It Actually Mean?
If I use AI to brainstorm ten ideas, choose one, rewrite it, fact-check it, and decide what it ultimately means, did a human make it? Most people would say yes.
If I type three words into an image generator and pick the best of twenty outputs, did I make it? That feels different.
The gap between those two examples shows up in more serious places than image generators.
An architect who uses AI to explore thousands of structural options but makes the final design call, and a doctor who uses AI to flag patterns in a scan but makes the diagnosis and talks to the patient, are both still making the human judgment that matters.
The tool changed, and human judgment remained critical to the output.
“Human-made” probably doesn’t survive as a single label once businesses actually have to use it. It’ll fragment into a vocabulary, the way “organic” split into certified tiers: human-authored, human-directed, human-verified, human-decided. A book can be human-authored while AI handled translation and cover design. A film can be human-made while containing AI-generated effects.
Collapsing each claim into one badge is where the current wave of certification runs into trouble.
The Strange Economics of Effort
For most of modern economic history, effort was relatively easy to infer from output. If something took weeks to produce, it probably involved a lot of work. If something was highly polished, someone probably spent time getting it there.
AI breaks that relationship. A highly polished piece of work might take three hours, or three minutes. A company can produce ten thousand variations of a message before lunch. The visible artifact no longer tells us much about the effort behind it.
That creates a strange economic question: if the machine can produce more, what exactly are we paying the human for? Not keystrokes. Not hours. Not volume.
We’re paying for judgment, context, taste, experience, accountability, intent. The ability to recognize that the technically correct answer is the wrong answer.
In other words, the things that are hardest to turn into output.
This Is Where Performance Gets Interesting
We’ve spent a lot of time measuring what people produce. That’s what scoreboards do. That’s what report cards do. That’s what performance reviews do. For a long time, that made sense, because output was a reasonable proxy for ability.
But AI is getting very good at producing the evidence. The essay. The presentation. The first draft. The thing we used to look at and say, “a person must know how to do this.” Maybe they do. Maybe they don’t. The artifact can’t tell us anymore.
If performance is increasingly easy to simulate, performance becomes a weaker signal of human contribution.
The question changes from what did you produce? to what did you contribute?
And eventually: what did you decide? That’s a much harder thing to measure.
Which is exactly why we’ll probably try to measure it anyway.
The Backlash Isn’t Really About AI. It’s About Care.
A growing backlash against AI-produced content makes it tempting to read it as a simple story: people don’t like AI.
I don’t think the evidence supports that.
In December 2025, McDonald’s Netherlands released a 45-second AI-generated Christmas ad. It was mocked immediately, called creepy, sloppy, and emotionally empty, and was made private three days after launch. The production company pushed back hard on the idea that this was a low-effort trick: CEO Melanie Bridge said the team hardly slept for seven weeks, generated thousands of takes, and then shaped them in the edit. “This wasn’t an AI trick,” she said. “It was a film.” The team says it spent substantial human effort getting the AI-generated material to work.
She may have been telling the truth about the hours. It didn’t matter. The audience wasn’t reacting to the actual labor behind the ad. They were reacting to what it looked like: something that didn’t feel as though anyone had cared enough to make it feel human. Actual effort and perceived effort had come apart, and the backlash followed the perception, not the timesheet.
A 2025 Journal of Business Research study, built on seven preregistered experiments, found the same pattern. Consumers who believe a message was written by AI rate it as less authentic and feel more moral disgust toward it, even when the message is word-for-word identical to a human-written one. But the effect is strongest for emotional communications, where you’d expect someone to have felt something, and it shrinks substantially when the AI only edited a human-written message rather than authoring it. People aren’t penalizing the technology uniformly. They’re penalizing its use where care was expected and didn’t show up.
The backlash isn’t about technology. It’s evidence that people are looking for a reason to believe someone cared, and right now the only tools they have for finding that reason are surface cues that don’t reliably track what actually happened, as McDonald’s own case shows. That’s exactly the gap a real provenance record is supposed to close.
The backlash tells us something important. People still want to know a skilled person was involved in producing quality work. They’re sick of what feels like “enshittification” in every area of their lives. Norway’s Consumer Council is even going viral for their campaign capturing what we all feel about this race to the bottom.
The Provenance Arms Race
Once a label becomes valuable, people find ways to game it. “Handmade” can mean many things. “Natural” can mean almost anything. Human-made won’t be immune. A company could put a human name on an AI-generated product and call it human-created, or have one employee make a token edit and claim oversight.
There’s a subtler, more sincere version already spreading through ordinary writing. The Wall Street Journal has reported on writers deliberately roughing up their own prose, dropping em dashes, breaking grammar rules, so clean writing doesn’t get mistaken for machine output.
We used to remove the fingerprints of the process. Now we’re adding them back, on purpose, because the fingerprints are what get believed. That’s exactly the kind of thing “human-made” labels will have to survive. If a label can be gamed by injecting deliberate typos or a rubber stamp, it stops meaning anything, the same way a self-applied “AI-free” badge with no verification behind it already does.
The label will be valuable because we trust what it means.
Human-Led, Machine-Assisted
Two different patterns show up.
In the first, AI quietly replaces the human. Not by decree, but by drift. It happens as we track output and prioritize volume and turnaround time. Judgment stops being the thing anyone measures, so it stops being the thing anyone protects.
The second pattern is the opposite overcorrection: treating the absence of AI as proof of human value. Some of these badges can be self-applied with no verification. But avoiding a tool doesn’t necessarily tell us who made the important decisions, how much judgment went into the work, or whether the result is any better. It tells us what tool wasn’t used. That’s not the same thing as proving what a human contributed.
Neither pattern holds up well on closer inspection. AI can legitimately do more of the production: draft the options, run the analysis, generate the variations, handle the routine. It still can’t decide what’s worth producing, why it matters, what good looks like, or whether a person is willing to put their name on the result and stand behind it when it’s wrong.
Those aren’t tasks that get automated away. They were always the actual job.
Human-made won’t always mean AI-free.
As we move towards the middle path, I predict we’ll land on human-led, machine-assisted: the machine did work, but a person decided what the work was for, judged whether it was good, and is answerable for it.
The label will show whether a human still mattered.
That’s a harder standard than a badge can currently prove, but it’s the right one. The point of provenance isn’t to romanticize human labor or pretend machines aren’t capable of remarkable things. The point is to say what actually happened: who contributed, what the machine did, what the human did, where judgment entered the process, who stands behind the result.
That last one may matter more than all the others. The argument isn’t about authorship. It’s about accountability.