Does using AI in creative production hurt a brand?

It hurts brands that use it without direction, because a general model's default output is the statistical middle of everything it has seen, and that middle is identical for you and your competitors. It helps brands that constrain it with an explicit system, anchor it to their own reference material, and keep a human accountable for what ships.

What should stay human?

The founding idea. The central concept of a campaign, the identity itself and the images a brand is remembered by come from a point of view, which is the one thing an averaging system cannot supply. The tools are then very good at multiplying that idea across every format and placement it needs to appear in.

How do you keep AI output on-brand at volume?

Three things together. An explicit system that states exclusions as clearly as inclusions, so it can be applied by someone who was not in the room. Reference-led production anchored to the brand's own archive rather than described from scratch. And a named human who reviews against the standard, because quality drifts downward invisibly when nobody owns it.

Should we disclose when work is AI-assisted?

You should have a written position before you are asked, covering disclosure, the usage rights on material feeding your system, consent for any synthesized likeness, and what never ships AI-made. The specific policy varies by category and audience. Deciding it calmly in advance is worth more than any particular answer.

Ten times the output, none of the voice

In short

Generative tools removed the cost of making more, which means volume is no longer a competitive advantage. What is scarce now is distinctiveness. Brands hold it by constraining AI production with an explicit system: a defined visual and verbal territory, reference material drawn from their own best work, and a named human who decides what ships.

Every brand we talk to has had the same year. The tools got good, the output went up, and somewhere in the middle of that the work stopped looking like theirs. Nobody decided this. It happened one deadline at a time, because the default output of a general model is, by construction, the median of everything it has seen.

That is the whole problem in one sentence. These models are averaging machines, and average is the opposite of a brand.

Why the output converges

A model asked for a product shot on a clean background will give you the most probable product shot on a clean background. So will the one your competitor is using, because it is the same model reaching for the same center of mass. The result is a category where everyone's work is technically competent, immediately produced, and completely interchangeable.

This is not a failure of the tools. It is what they are for. The failure is organizational: treating a general-purpose model as a creative department rather than as a production capability that needs direction.

A model asked for the obvious will give you the obvious, faster than you have ever received it before.

What actually holds a voice

Constraint, written down

A brand system that exists as taste in one person's head cannot survive contact with volume. It has to be explicit enough that it can be applied by someone who was not in the room: the palette and the deliberate exclusions, the lens and lighting behavior, the crop logic, the vocabulary you use and the vocabulary you refuse, the sentence rhythm.

The exclusions matter as much as the inclusions. Most brand guidelines describe what the brand is. Very few say what it must never do, which is exactly the guidance a production system needs when it is generating the hundredth variant at speed.

Your own work as the reference

The most reliable way to pull output away from the median is to anchor it to material that is already yours. Your best existing photography, your actual product, your real spaces and people, your archive. Reference-led production rather than description-led production is the difference between work that resembles your brand and work that continues it.

This is also where brands with real history have an advantage they routinely underuse. An archive is a moat.

A human who signs off

Volume fails at review, not at generation. If nobody owns the standard, the standard becomes whatever passed most recently, and quality drifts downward invisibly because each individual asset was fine. Someone has to be able to say no on aesthetic grounds and have that stick.

Where AI genuinely earns its place

We are not arguing against the tools. We build with them daily. The distinction we hold is between work where volume is the point and work where singularity is the point.

Variation of a resolved idea. The concept is set and directed, and the work is producing it across formats, ratios, languages and placements. This is where the tools are transformative.
Exploration before commitment. Fast, disposable visual thinking to find the idea, with nothing generated at that stage expected to ship.
The repetitive middle. Resizing, versioning, cleanup, assembly, first-pass QA. Unglamorous, expensive in human hours, and well suited to automation.
Not the founding image. The campaign's central idea, the identity, the photograph the brand is remembered by. Those still come from a person with a point of view, and the tools then multiply them.

The governance nobody wants to write

There is a boring layer under all of this that becomes urgent the first time something goes wrong. What is the disclosure standard when AI was involved. What are the usage rights on the material feeding the system. Whose likeness can be synthesized and with what consent. What never ships AI-made, as a written rule rather than an instinct.

Brands that write this down before they need it make a considered decision once. Brands that do not make it in a hurry, publicly, with a journalist waiting.

The test

Here is the only quality bar that matters, and it is easy to run. Put your last twenty assets next to your three closest competitors' last twenty, strip every logo, and ask someone outside the company to sort them by brand.

If they cannot, the volume gained nothing. You have paid to produce more of what everyone else already has.