WHY AI BRANDING CAN'T DO WHAT HUMANS DO

You can smell it before you can name it. The flyer in your feed, the logo on a startup deck, the "brand kit" someone generated over lunch. Something reads wrong, and it takes a second to figure out what. The gradient that means nothing. The typeface that almost exists but doesn't. The lighting from a render nobody asked for. AI branding has a look, and the look is the problem, because a brand that has a look instead of a system is not a brand yet.

We use AI tools every day at the studio. This isn't a complaint about technology. It's about people confusing image generation with brand building, and then wondering why the result feels like a stock photo of a company.

A logo is not an image, and AI only makes images

A logo has to survive conditions, not just look good once.

It has to hold at 16 pixels as a favicon, where everything but the silhouette disappears. Embroidered on a cap, where thin strokes vanish into thread. In one ink, in reverse, engraved, animated, cropped into an avatar, printed two meters wide. A designer is solving that entire range before the thing is ever attractive. Reduction, contrast, negative space, minimum legible scale, monochrome behavior.

AI solves one frame, because the screenshot is what it was trained to be rewarded for. Drop any generated logo to favicon size and it collapses into a smudge. There was never a structure underneath, only a pleasant arrangement of pixels at one resolution. Same with the file: generated marks are raster images pretending to be identities. Auto-trace one and the curves confess immediately, too many nodes, wobbly tangents, corners nobody decided. Anyone who has handed one of these to a printer knows the moment the file turns to sh*t and a human redraws the whole thing by hand anyway.

Average is the business model, not a bug

A generative model produces the statistical middle of everything it has seen. That's what it's for. Which is why AI branding across different tools, companies and industries all rhymes: soft geometric mark, cool gradient, clean sans, vaguely optimistic. The tools aren't bad at their job. They're excellent at it, and the job is producing the average.

Branding is the opposite job. Asking an averaging machine for distinction is like asking a thermometer to change the weather.

Now think about where the identities you actually remember came from. Someone chose an ugly color on purpose. Someone stretched a wordmark past comfortable. Someone looked at a category full of blue and went the other way knowing the client would hate it in the first meeting and love it in the third. Those moves are not averages, they're bets, and a bet requires somebody who can lose something. A model has nothing at stake. It will never risk being wrong, which also means it will never be interesting.

Design is a decision, and nothing here made a decision

Ask a designer why the blue. You get an answer: the three biggest competitors own warm palettes, the product needs to read precise rather than friendly, the color survives under retail fluorescents. A position you can argue with.

Ask a prompt why the blue. Silence. Something generated options until a human said "that one." Selection happened, reasoning never did, and reasoning is what the client is actually paying for. Six months later, when the marketing lead asks if the palette stretches to a sub-brand, nobody can answer, because nobody knows what the rules were.

Most identity work isn't producing, it's discarding. Killing the option that's prettier but says the wrong thing. That takes a thesis about the business, and a thesis comes from the room: what the founder wants but hasn't said out loud, the competitor keeping them up at night, the internal politics that will decide what actually ships, the face they make when they see the second option. Half our real input is read, not typed. None of it fits in a prompt, because the client doesn't know it well enough to write it down.

Then there's the part nobody can transfer. A designer brings twenty years of looking at things: the packaging from a childhood kitchen, a badly kerned sign on a street corner that stuck for no reason, an exhibition, a bad job, a client who once ruined a good idea. Taste is accumulated life, sorted into judgment. The reference a model pulls is an image it has seen. The reference a person pulls is an image they lived near, which is why it arrives attached to a reason.

Where AI genuinely earns its seat

We'd be lying if we said we don't use it. It's great at widening the field early: gathering references, generating fifty layouts to react against, cleaning up assets, drafting copy nobody will keep, killing the tedious middle. It gets us to the interesting problem faster.

What it can't do is choose. No stake, no argument, no reason to defend one direction over another. The value of a designer moved from producing to deciding, and that made judgment more expensive, not less. The studios hurting right now were selling production. The ones selling criteria are fine.

What the bad AI branding is actually telling us

The mediocre output is a free diagnostic. It shows us exactly what a brand with no thesis looks like, at scale, in high resolution.

Three things worth taking from it. If a two-line prompt can reproduce your brand, your positioning is generic and the visuals are just reporting that honestly. If your identity exists as a JPG and not as a set of rules, you have an image, not an identity. And if nobody on your team can explain why the logo is the way it is, there's no system to defend when the next campaign or hire arrives.

AI branding isn't killing design. It's just a very efficient mirror, and it keeps showing us the same thing: the work that survives was made by someone who had a reason, and was willing to be wrong about it.