AI Optimizes Toward Average. Brand Is How You Escape It.

You can generate a hundred versions of an ad before lunch. Different headlines, images, colors, calls to action, all of it, in an afternoon that used to take weeks. Agencies now build, test, learn, and optimize creative at a scale that was impractical a few years ago. This is real, it’s here, and it’s a genuine leap.

Let me be clear about how genuine, because what follows isn’t a case against AI.

The small stuff compounds.

In performance creative, tiny differences move real numbers. A different photograph. A sharper headline. A tweak to body copy, to brightness, to color, to the call to action. Any one of them can move a metric like click-through rate, and the winners aren’t always the ones you’d have bet on. The old constraint was that you could only make and test so many versions before the budget or the calendar ran out. AI removes that constraint. You generate the variants, put them in market, learn what works, and iterate, at a speed and volume that changes what’s possible.

Structure makes it better. Google’s ABCD framework, built with research partners like Ipsos, Nielsen, and Kantar, is a good example: Attract in the first few seconds, Brand early and throughout, Connect emotionally, give clear Direction. Google reports that following it drives roughly a 30 to 38% short-term sales lift and a 17% long-term brand lift. When I’ve applied those principles to video creative, engagement rose significantly. Pair a framework like that with AI’s ability to produce and test variations at scale, and you have a genuinely powerful optimization engine.

All of that is real. It’s also not the hard part.

Scale was never the scarce thing.

When making creative was expensive, more of it was the constraint, so that’s what everyone optimized for. Now output is nearly free. The bottleneck moved. The scarce input isn’t the ability to make the work. It’s knowing what the work should say.

Readers of the last issue know where this goes. The insight is the cargo. The execution is the vehicle. AI just gave everyone an infinite fleet of vehicles. It did nothing about the cargo. The best framework and the fastest test-and-learn loop in the world can optimize the expression of an idea. Neither can generate the idea worth expressing.

Optimization, left alone, drives toward average.

Here’s the part that should give you pause. Test-and-learn optimizes toward what performs right now. And what performs right now tends to be what’s safe, familiar, and proven. Run that loop long enough with nothing governing it but the metric, and it quietly erodes your brand’s distinctiveness. You optimize your way straight into the category average, because the average is exactly what tests well. Every competitor running the same loop lands in the same place. The tool that was supposed to sharpen your edge files it down.

Notice that Google’s own framework guards against this. Two of the four letters, Brand and Connect, aren’t about the metric. They’re about meaning and emotion, the human part. The most data-driven creative framework in the industry builds the brand and the feeling in on purpose, because optimizing for the click alone gets you forgettable. Even the optimizers know optimization isn’t enough.

The human job got harder, not smaller.

AI doesn’t replace the brand thinker. It raises the stakes on the one thing only a human supplies: the insight, and the brand truth that decides what the optimization is even allowed to chase. Let AI optimize the expression. Hold the line on the meaning. That’s the job now, and it’s a harder one than running the loop.

Part of holding the line is knowing what not to hand over. I have one rule. I won’t put an AI-generated face or voice in market. Not yet. They still read as synthetic, and the moment a person senses that a face or a voice isn’t real, something underneath the brand cracks. Not loudly. They won’t write in to complain. They’ll just feel that something is slightly off, and slightly off is corrosive to the one thing a brand can’t afford to lose, which is trust. The math is bad, too. The saving is small and the risk to the brand is large and hard to see. I’m not willing to make that trade for a face.

That’s a today line, not a forever one. The technology improves almost by the day, and at some point the uncanny edge disappears and this particular rule expires. The principle under it doesn’t: don’t put anything in market that makes a person doubt what’s real. The reason I won’t fake a face is the same reason AI can’t hand you the insight. The human part is the part people actually connect to.

Which is the whole point.

The insight is portable. It travels. A real human truth doesn’t belong to a category, which is why the best brand thinking crosses between them. The execution is the opposite: endlessly variable, specific to the channel and the moment, and now, increasingly, automated. AI transforms the part that was always going to vary. It can’t touch the part that was always the point.

So the threat was never AI coming for brand people. It’s that AI quietly removes the places people used to hide. When anyone can produce the work, production value stops being a differentiator. Volume stops being a differentiator. What’s left is knowing what’s worth making, and why it’s true. That was always the job. AI just made it the only job.

What’s the one thing you won’t hand to AI, and why? I’d like to know where other people are drawing the line.