Brand marketers are moving past the broad question of whether to use generative AI in creative work. The sharper question is where AI belongs in a customer-facing message, and where its presence changes what the audience thinks the brand is promising.
Recent examples from Koia, Hilton, Unilever, Goodwipes and Gozney show that there is no single brand-safe answer. Some teams are using AI to create deliberately unreal visual concepts. Others are refreshing existing photography or producing more paid-social variants. Premium brands are keeping the technology behind the scenes because visible synthetic content could weaken the credibility they sell.
The emerging playbook is less about a universal policy than a theory of appropriate use. AI creative works when the production method supports the meaning of the campaign. It becomes risky when efficiency starts rewriting the brand promise.
Table of contents
Jump to each section:
- What brands are testing in public
- Why context matters more than capability
- Efficiency gains are changing the creative portfolio
- The strategic tension between scale and distinctiveness
- What marketers should know about customer-facing AI
What brands are testing in public
Koia has used AI for concepts that are supposed to look larger than life, including animated or impossible product scenes. That choice makes the synthetic nature of the work part of the visual language, so the audience is not being asked to mistake an invented scene for documentary reality.
Hilton took a different route. Working with Shuttlerock and Meta, the hotel group used image-to-video technology to extend existing material across travel ads without returning to every property for another shoot. The use case was not to invent a new brand world. It was to make an established library work harder across performance media.
42% more reach, 8% ROAS and 13% more bookings were reported for Hilton’s AI-assisted travel ads, which also avoided new shoots across its property portfolio.
Goodwipes is also using AI to adapt visuals from existing photo shoots, while drawing a line at synthetic actors. For a personal-care brand, a generated spokesperson discussing an intimate product could create a trust problem that a modified background or reformatted image does not.
These examples point to a practical principle: the same technology carries different meaning depending on what the asset is claiming to represent.
Why context matters more than capability
A common assumption is that better models will make customer resistance fade. The contrasting reality is that technical quality does not settle whether an AI-made asset is appropriate. A flawless synthetic person can be more troubling than an obviously imaginary product scene because it makes a stronger claim about human presence.
That distinction explains why Gozney keeps AI out of consumer-facing work. The premium pizza-oven brand relies on craft, food culture and high-production storytelling to support a considered purchase. Synthetic imagery could make the work cheaper to produce while making the product feel less considered.
Context, not capability, sets the risk.
For marketers, the implication is to classify creative by the promise it carries. A functional product variation, an impossible fantasy sequence, a testimonial, a cultural story and a premium brand film should not pass through the same AI policy. Each asks the audience to believe something different.
Efficiency gains are changing the creative portfolio
The clearest commercial benefit of AI creative may be portfolio economics. Teams can refresh material from past shoots, produce channel-specific edits and support products that would never justify a new production budget. That expands the range of assets a brand can afford to test.
Unilever has used its AI Studio to accelerate asset production and create many variations from a smaller set of inputs. The strategic value is not simply faster output. It is the ability to move more products, occasions and audience segments into active creative rotation.
AI is making creative coverage cheaper before it makes creative judgment easier.
That shift can be especially important for smaller teams such as Koia and Goodwipes. They can compete for attention with concepts and iterations that once required a larger production operation. Yet the cost advantage only holds if the new output remains recognizable as the brand. More assets with less distinctiveness create inventory, not necessarily impact.
The strategic tension between scale and distinctiveness
Generative systems reward repeatable instructions, reusable templates and rapid variation. Brand building rewards recognizable choices, selective repetition and a point of view that competitors cannot reproduce. Those incentives overlap, but they are not identical.
The risk is not only visible failure, such as anatomical errors or awkward product details. A quieter risk is convergence. When many teams use similar models, prompts and optimization signals, technically competent work can begin to share the same visual logic.
Scale becomes strategically useful only when the system is scaling a brand decision, not substituting for one.
This is why brand guidelines need to evolve beyond approved colors, fonts and logos. AI creative governance should also specify which kinds of reality may be altered, when synthetic people are unacceptable, which product details must remain exact and what forms of disclosure fit the audience context. The goal is not to write a rule for every possible output. It is to preserve the decisions that make the brand legible.
What marketers should know about customer-facing AI
The most useful boundary is one that connects the production method to the claim an asset makes.
Separate fantasy from simulation. Deliberately impossible imagery can make AI part of the idea. Realistic synthetic people, places or product behavior carry a higher burden because audiences may read them as evidence.
Protect trust-sensitive moments. Testimonials, intimate categories, cultural representation and premium craft require more caution than background extensions or format adaptations.
Use AI to widen coverage. Existing photography, under-supported products and channel-specific variants are strong candidates when the underlying brand truth remains intact.
Govern sameness as well as errors. Quality control should ask whether an asset is accurate, but also whether it could plausibly belong to any competitor using the same tools.
The larger shift is that AI policy is becoming part of positioning. A brand that uses synthetic imagery openly for imaginative spectacle is making one choice. A premium brand that insists on photographed craft is making another. Both can be coherent if the production method reinforces the value the customer is buying.
Marketers will still need disclosure standards, approval paths and records of how assets were made. But process alone cannot answer the creative question. The durable advantage lies in knowing which parts of the brand can be automated, which can be transformed and which must remain visibly human.
As generative production becomes ordinary, restraint will become a form of differentiation. The strongest teams will not be the ones that use AI everywhere. They will be the ones that can explain why it belongs in this asset, for this audience, at this moment.