The Best AI We Use, You Never See

Last December, McDonald's Netherlands pulled an AI generated Christmas ad just days after release, once viewers piled on and branded it "AI slop." The studio behind it pushed back: its team had barely slept for seven weeks making it, generating thousands of takes and shaping them in the edit like any high craft production. "This wasn't an AI trick. It was a film," they said. Didn't matter. The ad wasn't bad because it was cheap or lazy. It was bad because everyone could tell.

That moment is the whole 2026 marketing story in miniature.

Consumer enthusiasm for AI is cratering: just 35% of Americans say they're excited about it, against 80% who are concerned. Mentions of "AI slop" are up ninefold in media monitoring. Seven in ten people say AI made ads feel like they're missing something. And here's the line that should stop every founder cold: per a 2026 Gartner survey, half of US consumers would rather buy from brands that don't use generative AI in their customer-facing content at all.

Now hold that next to this: 99% of marketing leaders plan to increase AI spend in 2026, and 97% already use it in their daily work.

So the entire industry is flooring the accelerator on the exact thing a growing chunk of customers is learning to resent. That's not a strategy. That's a stampede.

There are really only two places you can put AI in a business.

In the output: the copy, the creative, the video, the customer-facing post. This is where the race is, and it's also where the trust penalty lives. 78% of consumers say they'd rather see ads made by people even if AI could make them better. Read that twice. Even if it's better. The robot smell is its own deduction.

In the work: the research, the targeting, the timing, the grunt work, the tenth draft you'd never have had time to write. This is invisible to the customer. And it's where almost all the real advantage is hiding, precisely because everyone's too busy generating spokesmodels to bother.

Our whole bet at Propello fits on a bumper sticker: pour AI into the engine, keep it out of the storefront.

Last issue we called the "AI layoffs" what they mostly were: cowardice with a deck. Two weeks later, some of the biggest names who'd predicted a white-collar wipeout started saying the same thing.

At a Sydney event on May 26, OpenAI's Sam Altman admitted he'd been "pretty wrong" about AI's impact on jobs. The entry-level wipeout he forecast simply hasn't shown up. Anthropic's Dario Amodei, who once put half of all entry-level white-collar jobs on the chopping block, now frames automation as something that expands what people do. And Nvidia's Jensen Huang went furthest, calling the AI causes layoffs narrative "too lazy" and saying the industry was scaring people for no good reason. The tech barely arrived, so how is it already eating jobs?

The timing isn't subtle. Both OpenAI and Anthropic are walking toward IPOs in the neighborhood of a trillion dollars, and "our product will vaporize the workforce" turns out to be a hard sell to pension funds. Meanwhile the Yale Budget Lab found no meaningful unemployment bump for high AI exposure jobs through March 2026.

The lesson hasn't changed since last issue. The model is leverage. The human is the business.

AI is everywhere in our stack, and a human decides everything that matters. Walk it across our four phases:

Define. Research that used to eat a week eats an afternoon. AI maps the landscape, surfaces the competitors, drafts the questions. Then a human looks at ten plausible problems and decides which one is worth your runway. The model narrows the search. It does not place the bet.

Design. AI lets us explore ten creative directions in the time it used to take to explore one. More shots on goal. But "which of these ten should a brand stake its identity on" is a taste question, and taste is not a feature you can prompt for. A human picks the one that's actually you.

Develop. Claude writes alongside our engineers and collapses the boilerplate so they live in the 20% where senior judgment actually pays: architecture, what to build, what to kill. (Reminder from last issue: when developers outsourced that judgment in the METR trial, they felt 20% faster and were measurably 19% slower. Leverage in the right hands; an anchor in the wrong ones.)

Launch. Here's the marketing punchline. We run AI hard on the invisible engine: segmentation, send time, audience modeling, A/B variants, repurposing one good idea into twelve formats, reading what's converting in close to real time instead of waiting a month for a report to confirm what already happened. The customer never sees any of it. What they do see (the voice, the hook, the thing that's supposed to make them feel something) stays human. Because 87% of people will tell you the best advertising still needs a human touch, and they vote with their wallets.

AI in your marketing should work like a great session musician: indispensable in the studio, invisible on the record. The second the audience notices the machine, you've already lost the thing you were trying to build.

So stop trying to make AI your face. Make it your back office. Use it to ship the campaign that was too expensive last quarter, to test the twelve angles you didn't have hours for, to know your customer better than your competitor does. Then put a human on the microphone.

Everyone can rent the same model. Few are careful enough to keep it off the stage.

A couple of weeks ago we launched WWFTW: a one minute, two question study asking people what they think would actually fix the world. We're collecting answers globally, clustering the themes, and running simulations on where the most common ideas would lead.

Most of you opened that email, nodded, and meant to come back to it. This is your nudge. It really is sixty seconds.

Add your answer →

Methodology and what we're building next: propellostudio.com/what-will-fix-the-world

P.S. We're building this AI in the work series straight out of our own client projects: engineering done, marketing today, design and GTM still to come. If there's a function you want us to pull the curtain back on, just hit reply.

Abhishek The Velocity Digest · Propello Studio