Provenance Is Not a Business Model

There is a number going around my industry that feels like good news, and I think it is being read exactly backwards. Seventy-eight per cent of consumers say they would rather see advertising made by people, even if AI could produce something technically better. It comes from Canva's 2026 study with The Harris Poll — 3,547 consumers and 1,415 marketing leaders across seven countries — and as survey work goes it is solidly built.

It is also, on its own, worth very little to anyone trying to make a living. I want to explain why, because the conclusion I draw from it is not the one most people in my field draw, and I would rather argue the case than imply it.

The question that was not asked

Read the item carefully. People were asked which they would rather see. They were not asked what they would pay, what they would give up, or what they would choose when the human version costs three times more and arrives a week later. There is no willingness-to-pay question in the study at all.

This is not a criticism of the researchers. It is a normal preference question and it measures a real thing. But a stated preference with no price attached is not yet a market. Almost everyone says they prefer food made from scratch, furniture built by hand, and clothes that last a decade. The revealed behaviour of those same populations built the industries that supply the opposite. The gap between what people say they value and what they reliably pay for is one of the most durable findings in consumer research, and there is no reason to expect creative provenance to be the exception.

A penalty is not a premium

Here is where the argument usually goes wrong, including in pieces I have read from people I respect.

There is real experimental evidence that labelling something as AI-generated hurts it. Across a number of controlled studies, telling people that a piece of marketing was machine-written lowers how authentic it feels, lowers engagement, and lowers stated purchase intent. That effect appears to be genuine and it is measured under conditions where you can actually attribute cause.

But notice what that is. It is a penalty for disclosure. It is not a premium for certification. "People think less of your work when they learn a machine made it" and "people will pay more for a certificate saying a human made it" are different propositions, and only the first one has evidence behind it.

I went looking for the second. I wanted to find a brand that had charged a documented, sustained premium for verified human authorship and could show the revenue. What I found was trend pieces about the idea of a human-made premium, several competing certification badges, and a good deal of confident writing about where the market is heading. I did not find transactions. That is not proof that none exist — an absence I searched for is a weaker claim than a fact I verified, and I am stating it as the weaker claim. But if the premium were working at scale, the case studies would be the easiest thing in the world to find, and they are not.

What the budgets actually did

Meanwhile the money has a direction, and it is measurable.

Forrester's 2026 B2B Brand and Communications Survey reports that the share of B2B marketers expecting to increase agency spend on content creation fell from 41 per cent to 26 per cent year over year. For digital marketing services the same measure fell from 51 per cent to 31 per cent. I want to be precise about what that is, because it gets quoted loosely: it is the proportion of buyers who expect to spend more, not the total amount spent. It is an intent measure, and intent measures move faster than budgets. But a drop of that size in one year, in both lines at once, is not noise.

On the supply side the evidence is real in direction and genuinely thin in construction, and I am going to say so rather than use it and hope nobody checks. Creative Boom's July 2026 survey puts the median UK illustrator day rate at £350, with the middle half between £280 and £450. That median rests on ninety usable answers from UK respondents. Ninety. I refused to use a widely quoted statistic three weeks ago partly because it surveyed around two hundred and thirty self-selected readers of a single publication, and I am not going to pretend a smaller sample from a single publication is sturdier because its conclusion suits my argument. Around sixty per cent of the same respondents said AI had affected their work or income in the previous twelve months, and about a third of those said significantly.

The Association of Illustrators has a much larger number — 6,844 respondents, of whom more than thirty-two per cent reported losing commissions to generative AI, at an average estimated loss over nine thousand pounds. That sample is far bigger, and it was gathered by an advocacy organisation as part of a response to a government consultation about AI harm, which is close to the ideal conditions for attracting people who have been harmed. Both of these numbers point the same way. Neither of them is strong enough to be the load-bearing wall of an argument, and I would rather build the argument on the buyer-side data, which is.

Two things that are both true

Here is the part that I think gets lost, and it is the reason this essay is not simply pessimistic.

The strategic value of being distinctive is rising. When the cost of producing competent, unremarkable work falls to nearly zero, competent and unremarkable stops being a position anyone can defend. Everything that reads as average becomes commodity almost immediately, and the only work that retains pricing power is work that could not have come from anywhere else.

The market price of provenance is falling at the same time. Those are not in tension. Being distinctive is a property of the output — of a point of view, a standard, an obsession, a body of work that accumulates into something recognisable. Provenance is a fact about the production process. Buyers pay for the first because it changes what they get. They do not pay for the second because it does not.

Which is why "made by a human" is a poor thing to sell, and why I think a good deal of the current positioning around it is going to disappoint the people adopting it. It asks the buyer to care about your process. Almost nobody has ever paid extra to care about someone else's process. They pay for a result they could not obtain elsewhere, or for a risk they no longer have to carry.

What I would do instead

Sell the outcome and the standard, not the origin. Be specific enough about what you will and will not do that the specificity itself becomes hard to copy. Publish the method so that the quality claim is checkable rather than asserted — that is what I try to do here, and it is a commercial argument as much as an ethical one. Compete where the evidence says buyers actually spend, which is on distinctiveness and on results, and stop competing on a fact about production that nobody has been shown to pay for.

If someone produces a case study — a real one, with revenue attached — showing a brand sustaining a price premium for certified human authorship, I will publish it and change my position. I have looked, and I would genuinely like to be wrong about this one.