Why I Pulled My AI Ads (and What the Data Says)

Hey Everyone - Hope you had a great weekend. I spent mine on a pickleball court trying to undo a year of sitting at a desk (more on that below). I have also been turning over an experiment I ran on my own newsletter business this quarter that did not go the way I expected, and a stack of new consumer survey data that explains why.

This week:

Let's dive in.

In December, McDonald's Netherlands pulled an AI-generated Christmas ad after a wave of public backlash. The comments did not focus on craft. They focused on the message the ad sent: "we optimized for production cost over the human work that might have actually meant something to you during a season that matters." Coca-Cola took similar heat for a similar reason. Both companies have global marketing budgets capable of producing traditional ads. They chose AI anyway. Five months later, we have the data to explain what they paid for that choice.

Three months ago, in the post "When Will Non-AI Goods Become Premium?", I argued that synthetic abundance would push human-made into the premium tier, the way mass production pushed farmers markets into the premium tier. I was making a prediction. The numbers are now in.

The authenticity premium is no longer a niche luxury position. It is the median consumer preference, and it is forming faster than almost any marketing trend I have tracked.

I lived this myself this quarter. I had been running AI-generated ads for one of my newsletters. They worked. The economics were good. I was happy to keep going. Then the complaints started. Not many. Maybe a few dozen out of a much larger audience. But the readers complaining were not random. They were exactly the readers I most wanted to keep. They cared enough to write back. The cost of losing them was higher than the savings of producing the ads cheaply, so I stopped.

That experience surfaced a question worth answering directly, because it is the most interesting wrinkle in this entire pattern.

What happens when AI gets so good that nobody can tell?

Here is what I think is going on, and what I am betting on.

First, detection is not the only thing keeping the authenticity premium alive. The premium is also propped up by the cultural prior. Once a critical mass of consumers learns that AI content is everywhere, the default assumption about every piece of content shifts. Every ad starts paying a small skepticism tax, whether or not it is actually AI. Economists call this a market for lemons. When buyers cannot tell honest sellers from dishonest ones, average prices collapse and honest sellers exit. Advertising is on its way to that kind of market, and a more capable AI does not fix it. A more capable AI accelerates it, because the suspicion is the same whether the AI is good or bad.

Second, when consumers cannot trust the artifact, they shift their trust upstream. They stop trusting the ad and start trusting the source. The brand with a decade of consistent values becomes the signal. The creator with a recognizable voice becomes the signal. The friend who recommends becomes the signal. This is good news for anyone who has invested in real relationships and a verifiable track record. It is bad news for anyone whose only asset was the ability to put attractive content in front of strangers.

Third, the only ad signals that will keep working in an undetectable-AI world are the ones that are physically expensive to fake. A real founder on camera, talking about a real moment in their company. A photograph with a date and a location in it. A handwritten note inside the package. A field recording with ambient sound. A live event you can drive to. These are not better because they look better. They are better because the cost of faking them is high enough to be a credible signal of human stakes.

So the answer to "will ads just stop working" is both yes and no.

Generic ads, generic copy, generic creator content. These will keep getting cheaper to produce and will keep working less and less, because the cultural prior is moving against them faster than the technology is moving toward undetectability.

Specific ads, with a verifiable person attached, with friction-laden evidence of human craft, placed in a context where the audience already trusts the source. These will work better than ever. The premium will move from the artifact to the seal of the person putting it out.

This is the frame I want you to carry into your week. When you look at your customer-facing surfaces, do not ask "is this good AI or bad AI." Ask these four questions.

For every surface that hits any of those, AI is subtracting value, even when it is saving you time. The fix is not to write the same copy with a different tool. The fix is to put something on that surface that an AI demonstrably could not have produced. A name. A photograph. A specific story. A receipt.

McDonald's paid the authenticity premium in a single news cycle. The brands and creators that will compound through this decade are the ones paying it on purpose, every week, before the headline forces them to.

A few of you know I played golf in college, pro for a (very short) stretch, and a fair amount even after I stopped chasing it as a career. What I miss most about that life is not the game itself. It is what the game did for the four hours around it. Golf is a four-hour walking conversation with a built-in excuse to be outside. You get to know somebody differently when you have watched them three-putt under pressure, or laugh at their own bad shot, or be patient with a slow group ahead.

I have been thinking a lot about this, and the conclusion I keep landing on is that connection happens when you do stuff together. Not when you sit across from each other at a table. Not when you text. When you are pointed at the same thing, with your bodies in the same place, and there is some friction that you both have to work through. That is the unfakeable substrate of a relationship.

So this week I am building a new medium for that. I am playing pickleball.

The choice is partly opportunistic. Pickleball is everywhere in Austin right now, the entry barrier is laughably low compared to golf, and you can get a doubles game going with three texts and a half-empty rec center. It is also partly that I have something to fix. My back has been giving me trouble. I have gained more weight than I want to admit in print. Sitting in front of a screen building newsletters and running ad funnels does not bend or stretch you in any of the directions a body actually needs.

I keep coming back to the fact that my body is one of the few things in this world I have real, daily, unambiguous control over. The market is going to do what the market does. AI is going to compress whatever it is going to compress. Most of the levers I touch are coupled to other people's decisions. My body is not. If I eat well and I move regularly and I sleep, my body responds. If I do not, it responds the other way. That feedback loop is honest in a way that almost nothing else in my life is.

So the build this week is dumb on purpose. Show up to play. Twice a week, minimum. Sweat a little. Get to know whoever else is there. Notice the back. Be patient with the back. Lose to people I should beat. Eat a real dinner afterward.

If you want the meta point, this is the same hedge as the farm, the same hedge as the Founders Feed, the same hedge as cooking for people. The world is moving toward more screens, more synthetic content, more outsourced thought. The counter-move is to keep practicing the things that do not scale, that require my body and presence, and that put me in rooms with other people doing the same. Pickleball is just my newest version of that counter-move.

If you are in Austin and want to play, hit reply.

Things I Learned:

Every product, every service, every piece of content benefits from a deliberate human fingerprint. A handwritten note tucked inside the package. A voice memo instead of a typed email. A first-person photo of you holding the thing. The credits at the end of the video naming the people who actually made it. A live phone number, your own, that a real person picks up.

This is what the luxury houses have already started doing, with "no AI was used" callouts and visible craft inserts in their campaigns. You do not need a Burberry budget to copy the principle. Pick one customer-facing surface this week and add one thing nobody else in your category is putting there. A name. A face. A timestamp. A specific story that nobody could have hallucinated. A receipt of the work.

I do this for in person events I go to. I got into making trinkets with my fiber laser a few years ago, and when I meet new people I often bring them a rock or a magnetic name tag that I made. Surprisingly, no one has thought that it was super weird (or at least they haven’t told me that it was).

The reason this works is not because it looks better. It works because the cost of faking it is high enough to be a credible signal. AI can generate the appearance of effort. It cannot generate the fact that a specific person made a specific choice on a specific day. That fact, made visible, is the moat.

This skill is durable regardless of how the AI trajectory plays out. If models stay flawed, your fingerprint stands out by contrast. If models become flawless, your fingerprint stands out because it is the only thing the model could not produce. Either way, the handmade signature is appreciating in value while AI output deflates.