What happens when the decision-maker never actually sees your ad?
“Looking great” is no longer a single thing.
It used to be. A brand published a visual, a campaign image, a product shot, a banner ad, and the question was simple: Does this stop someone? Does it hold attention? Does it make someone want something? Human eyes, human emotions, human decisions. The entire discipline of visual communication was built around that straightforward transaction.
That transaction still exists. But it will very soon share space with another one that most brands are not designing for.
Very soon, the entity encountering your visual content will not be a person. It will be an agent, deployed by a consumer to research, compare, evaluate, and recommend; a shopping agent assessing product options, a travel agent comparing hotel brands, a service agent vetting vendors. These agents act on behalf of humans, report back to them, and, in many cases, make decisions that humans simply ratify. The human may never see your image at all.
Which raises a question the visual communications industry has not yet fully confronted: what does your content mean to something that processes rather than feels it?
We have been here before — but not like this
The challenge of serving two audiences simultaneously is not new. Brands learned it with search. When Google’s crawlers began indexing the web, text content had a visible layer for humans and a structural layer underneath: keywords, link architecture, semantic markup, and metadata. SEO was the discipline of making those two layers work together without sacrificing either. It became fluent, then invisible. It is now simply part of the process of making content.
What is new is that this same dual-layer reality has arrived for visual content, and the visual industry has no equivalent fluency yet. There is no Visual SEO for AI agents. No established practice for making a brand image legible to a non-human audience, much less authoritative.. The tools, the vocabulary, the workflow integration: none of it exists at scale.
And the stakes are different this time, because the agent is not just indexing. It is deciding.
What agents actually see and how they see it
The first instinct is to assume agents simply read images poorly. The research suggests something more nuanced.
Modern multimodal models read visual content with considerable accuracy. A 2025 Royal Society study found that AI emotional ratings on visual scenes correlate strongly with average human ratings, across valence, arousal, and discrete emotions, including happiness, fear, and sadness, with correlation coefficients ranging from 0.77 to 0.91. The interpretation is often similar. The difference is in everything that comes after.
There is so much that alters how an emotion impacts a decision. The weather, the time of day, one’s health, one’s personal memories, the surrounding noise, the amount of sleep, the amount of coffee previously ingested… the list is almost endless.
A human who feels something from an image is moved toward or away, consciously or not. An agent that classifies the same image as emotionally positive routes it accordingly, but the routing logic is functional rather than felt. The classification may be accurate. The experience is absent.
For brand communication built on emotional connection, the specific quality of resonance that makes someone choose one thing over another- that gap matters enormously. The classification will always be an average assessment.
A human reads cultural conventions. An agent reads a visual signal.
There is also a precision asymmetry that runs opposite to what most people expect. Psychologist David Navon established in 1977 that humans see the forest before the trees — we respond to global features first, and local details only insofar as the overall impression allows them through. A small inconsistency, a slightly off-brand detail, a compositional choice that sits outside the brand’s visual language: a human absorbs these into the overall impression and moves on. An agent processes without that filter. Everything registers with equal weight. None of it gets smoothed over. It all counts.
This is also where certain creative strategies become fragile. Research on visual juxtaposition shows that AI still struggles significantly with placing contrasting elements side by side to create meaning through contradiction, with three compounding challenges: accurate interpretation of individual elements, comparative reasoning across those elements, and the social and cultural comprehension that makes the whole thing legible as intended. A campaign whose intelligence lives in the gap between what is shown and what is meant- irony, subversion, and deliberate tension- is precisely the creative most at risk of being misread at the machine layer.
Consider a luxury fragrance campaign built on deliberate restraint — no product visible, no aspirational setting, plain clothing, grey light. To a human viewer fluent in the visual grammar of luxury advertising, the subversion is the signal. Restraint communicates confidence. The refusal of convention communicates that the brand is beyond it. An agent sees low saturation, absent product placement, and no recognizable category cues. It does not experience the subversion. It sees the absence.
Finally, AI emulates an average human response. Brand creative that works because it speaks to a specific subculture, a generational reference, a niche identity, content whose power is in who it resonates with, rather than how broadly, may register as merely adequate to an agent seeing only the statistical center. The edges of the audience, where brand loyalty is often built, are where machine evaluation is least reliable.
Beyond alt-text
When agents cannot fully derive meaning from the visual itself, they read what surrounds it. Structured metadata, product descriptions, and schema markup become the primary signal. A beautifully crafted advertisement may be functionally invisible to the agents deciding within its category. The art direction, the lighting, the compositional intention: none of it registers. The agent reads the price, the rating, and the structured attributes. The creative investment evaporates at the machine layer.
This makes metadata critical, not as a replacement for strong visuals, but as their translation. The data surrounding an image must carry the meaning the image was designed to convey. Not “blue jacket, model, outdoor setting,” but the intent, context, tone, and positioning a human would instantly extract from the visual. Expressive metadata, not merely descriptive metadata. How you transform a compelling advertisement into an efficient one at this layer is a skill the industry has yet to develop.
What this means now
The question is not whether to choose between human resonance and machine legibility. It is whether brands develop the awareness to pursue both, not as a compromise, but as two distinct creative acts that must remain coherent with each other.
That means metadata that carries meaning, not just description. Creative review that asks what an agent understands alongside what a human feels. An honest reckoning with the kinds of creativity that depend entirely on human cultural reading, irony, niche resonance, subversion, and what it means that this layer of the work is currently invisible to the machines routing it.
None of this has a standardized practice yet. But the shape of the problem is clear enough to start building toward it.
The discomfort worth sitting with
We did not take search seriously until rankings mattered. We may not take this seriously until a generation of creative investment has been performing for an audience that was never there.
This article was written for Santa Cruz Software and was first published in DAM News . Main photo : Photo by Joshua Rawson-Harris on Unsplash
Author: Paul Melcher
Paul Melcher is a highly influential and visionary leader in visual tech, with 20+ years of experience in licensing, tech innovation, and entrepreneurship. He is the Managing Director of MelcherSystem and has held executive roles at Corbis, Gamma Press, Stipple, and more. Melcher received a Digital Media Licensing Association Award and has been named among the “100 most influential individuals in American photography”
Comments are closed, but trackbacks and pingbacks are open.