The Hidden Cost of Designing for the Average

May 24, 2026

10 min read

1 views

61%

What happens when the people inside the margins are also left out of the design?

Most systems fail in the same familiar way: they are built for the imagined average user, then patched later for everyone else. But the deeper failure is not just exclusion at the edges. It is what happens when a person is already inside a marginalized group and still does not fit the default inside that group. That is where invisibility compounds, and where the cost of bad design becomes hardest to see.

Now consider a different kind of system failure. A tool appears that can generate a UI quickly, maybe even from a prompt, and can be edited inside Figma. Suddenly the bottleneck of interface creation feels smaller. A solo founder, who once needed a designer, a developer, and weeks of iteration, can now produce a polished product in a fraction of the time. That sounds like liberation. It is also a test.

The test is this: what kind of users will be visible to these systems, and which ones will disappear even faster than before?

The connection between intersectional marginalization and AI assisted design is not obvious at first. One lives in the social structure of identity, the other in the workflow of product creation. Yet they share a brutal similarity: both are shaped by assumptions about the norm. When those assumptions go unchallenged, the people who deviate from the norm are not merely underserved. They become structurally unreadable.

That is the real tension. The more powerful our systems become, the easier it is to mistake speed for understanding.

The average user is a myth, but a profitable one

Every design system has an invisible center. It imagines a user whose needs, body, language, behavior, and context are stable enough to organize everything around. This is efficient. It reduces complexity. It also creates a quiet violence: anyone who does not match the center must explain themselves repeatedly.

A wheelchair user on a website is not just a wheelchair user. She may also be a Black woman, a parent, a bilingual speaker, a trans person, a low income customer, a person with low vision, or some combination of these. If a product is “accessible” in one narrow sense but fails in another, that user is still forced to work around the system. The problem is not only the missing feature. It is the assumption that one axis of inclusion is enough.

This is why intersectionality is not a niche social theory but a design principle. It reveals that exclusion is often cumulative rather than additive. A person does not experience racism plus sexism plus transphobia as separate lines on a checklist. They experience the compound effect of being seen through overlapping filters, each one changing how the next is interpreted.

The same thing happens in products. A form may be technically usable, but it assumes a legal name, a single first name, a standard address, a conventional family structure, and a flawless internet connection. Each assumption seems small. Together they define who belongs.

The average user is not a real user. It is a compression artifact, a way of flattening human variation into something easier to optimize for.

That flattening is especially dangerous when tools promise to make creation easier. The faster you can generate an interface, the more likely you are to reproduce the defaults embedded in your prompt, your template, or your training data. Speed does not eliminate bias. It scales it.

Visibility is not the same as inclusion

One of the most seductive features of modern design tools is that they make output visible immediately. You can see the screen, click the button, move the card, change the color, and feel progress. But visibility can be deceptive. A thing can be visible and still not be included in the logic of the system.

This is true socially and technically. A marginalized group may be acknowledged in language, represented in marketing, or mentioned in policy, while the actual structures continue to fail the people most at risk. Likewise, a UI can look clean and polished while excluding the users who encounter the most friction. A beautiful interface is not the same as a hospitable one.

Think of a restaurant with a stylish front door and no ramp. The place is visible. The brand is visible. The menu is visible. But for some people, the restaurant is still unreachable. Now imagine that same mistake repeated in digital form: a checkout flow that breaks for screen readers, a identity form that rejects chosen names, a support chatbot that cannot interpret slang or code switching, a prompt generated interface that silently assumes Western conventions.

The deeper issue is that visibility often flatters the designer. If the product looks coherent, it feels complete. If the interface is editable in Figma, it feels controllable. If the generation is fast, it feels intelligent. But none of those qualities guarantee that the system can recognize the user who exists at the intersection of multiple exclusions.

This is where the analogy to marginalized within marginalized groups becomes powerful. The challenge is not merely whether a group is seen. It is whether the subgroup inside that group is seen as anything other than an edge case. The Black trans woman, for example, is not simply a member of two categories. She may be erased within movements that celebrate one identity while ignoring another. In product terms, this is the person whom your accessibility checklist misses because she is both the kind of user you claim to support and the kind your assumptions were never built around.

The lesson is uncomfortable: inclusion that stops at the most legible case is not inclusion, it is partial visibility with a marketing budget.

AI makes the old blind spots faster, not automatically worse or better

When people say a tool is like ChatGPT for UI design, they usually mean productivity. Less friction. Faster mockups. More autonomy for solo builders. Those are real gains. A solo entrepreneur can now explore ideas that once required a full team. A designer can iterate before the first meeting. A founder can test a concept without waiting for resources that may never arrive.

But every acceleration tool changes the shape of what gets noticed. The danger is not that AI creates entirely new bias from nothing. The danger is that it compresses the feedback loop around preexisting assumptions. If your prompt asks for a “modern fintech dashboard,” the system will happily return a polished consensus version of what modern fintech is supposed to look like. If your own mental model of the user is narrow, the tool will give that narrowness a professional finish.

That matters because design often fails at the level of imagination before it fails at the level of implementation. If no one imagines that a user may have two surnames, a caregiver role, limited literacy, a disability, a queer family structure, or a nonstandard relationship to institutions, the resulting product may still appear elegant. The exclusions are simply hidden under the surface.

A useful mental model here is the three layers of design blindness:

Template blindness: the system starts from a default shape that privileges some users over others.
Prompt blindness: the creator asks for what they already know, so the tool reproduces their worldview.
Review blindness: everyone checks whether the interface looks good, but no one checks whether it works for people at the intersections.

AI can reduce the cost of making, but it does not reduce the cost of not thinking. In fact, it can increase that cost by making negligence scalable.

This is why the most important question is not whether a tool can generate a UI. It is whether the tool helps creators notice the users they would otherwise overlook. If it does not, then it is a speed layer on top of a narrow imagination.

The real design challenge is to build for compound lives

Human beings do not arrive in single categories. They arrive as stacks of lived realities. A person may be a student and a caregiver, disabled and ambitious, religious and queer, undocumented and highly skilled, elderly and digitally fluent, or any number of combinations that no persona deck fully captures. Good systems do not just accommodate difference. They expect it.

This is where a more ambitious design philosophy begins. Instead of asking, “How do we make this usable for the average person plus a few exceptions?” ask, “How do we make this resilient for compound lives?” That shift matters because compound lives are not edge cases. They are the standard condition of actual humanity.

Consider a loan application. A simplistic design might assume a fixed address, a linear employment history, and one official name. A better design recognizes that people move, work irregularly, change names, share housing, and live under constraints that do not fit neat corporate categories. The good system does not force the person to become legible by flattening their life. It expands its own grammar.

Or take a healthcare portal. A narrow design may ask for “mother” and “father,” “male” or “female,” a singular emergency contact, and a binary relationship model. A compound life aware design allows for chosen family, multilingual navigation, alternate identities, privacy concerns, and multiple pathways through the same workflow. The difference is not cosmetic. It determines whether the user can participate at all.

This is also the point at which speed and justice can be reconciled, but only partially. AI assisted design can help teams explore more variants faster, which is valuable. Yet the real opportunity is not simply producing more screens. It is using that speed to prototype for the users who are typically hardest to imagine, then validating with them early enough to change the system.

The best workflow is not prompt, generate, ship. It is prompt, generate, interrogate, revise.

The goal is not to make design faster than thought. The goal is to make thought fast enough to keep up with design.

What responsible creators should do next

If intersectional marginalization teaches us anything, it is that omission is rarely random. It usually follows patterns of power. If AI assisted design teaches us anything, it is that those patterns can be replicated at unprecedented speed. Put together, the two ideas create a simple but demanding mandate: do not mistake convenience for coverage.

That means creators need a new checklist, one that is less about polish and more about epistemic humility. Before celebrating a fast UI, ask who was assumed in its making. Before declaring a product accessible, ask which barriers were tested and which were invisible because no one on the team had lived them. Before shipping a polished flow, ask whether someone at the intersection of multiple exclusions could actually move through it without translating themselves into a simpler identity.

It also means broadening who gets to review the output. The best validator is not always the most technical person in the room. Sometimes it is the user who routinely has to navigate forms, policies, and interfaces that were never built with their life in mind. Their corrections are not edge feedback. They are structural intelligence.

There is a moral and practical reason for this. The systems that forget compound lives do not merely fail a minority. They produce brittle products. Any design that only works for people who live in clean categories will eventually break in the messy reality where most customers actually live.

Key Takeaways

Stop designing for the average user. The average user is a statistical fiction that hides real human variation.
Treat intersectionality as a usability framework. If a product fails users at the overlap of identities, it is not fully inclusive.
Use AI to widen imagination, not to replace it. Fast generation should create more room for exploration, testing, and correction.
Review for compound lives, not just single-axis accessibility. Ask whether the system works for people whose needs stack, collide, or change over time.
Assume every default is a political choice. Any field, label, workflow, or template may quietly decide who counts and who has to adapt.

The future belongs to systems that can recognize complexity without flattening it

The most interesting question is not whether AI will replace designers or whether marginalized people will be better represented in products. It is whether our tools will teach us to see complexity more clearly, or whether they will make it easier to ignore complexity at scale.

A UI generator can produce a screen in seconds. A just design culture takes much longer. It requires learning that a person is not fully described by any single identity, and that a polished interface can still be a locked door. It requires understanding that the people most likely to be overlooked are often the ones with the most to teach us about what systems are missing.

So the next time a tool promises to make design easier, ask a harder question: easier for whom, and visible to whom? The future will not be defined by who can generate the fastest interface. It will be defined by who can build systems that remain humane when the user does not fit the template.

That is the real frontier. Not just more design. More recognition.