Why Do AI-Generated Algorithmic Interfaces Feel Wrong?

An interface can perform well according to analytics and still feel uncomfortable to use. People may complete tasks quickly, click the intended buttons and follow the expected paths, yet describe the experience as confusing, impersonal or difficult to trust.

This apparent contradiction highlights a limitation in how digital products are often evaluated. Metrics such as conversion, click-through rate, engagement and task completion can show what users did, but they do not fully explain how the experience felt.

As artificial intelligence becomes more involved in generating, personalising and optimising interfaces, this distinction becomes increasingly important. Mathematical efficiency and psychological usability are related, but they are not the same thing. In some cases, improving one may weaken the other.

The Illusion of Optimisation

Algorithmic systems are usually optimised against measurable objectives. Depending on the product, these might include:

• increasing purchases or subscriptions
• reducing the time needed to complete a task
• encouraging users to view more content
• increasing interaction with particular features
• improving the visibility of recommended actions

These measures are useful, but they are only partial indicators of user experience.

A user may complete a task quickly because the interface strongly directs them towards a particular action. This does not necessarily mean that they felt informed, confident or in control. Similarly, an increase in engagement may result from relevance, but it may also come from distraction, repetition or difficulty finding a clear stopping point.

Human-centred evaluation therefore needs to consider more than system performance. Research into explainable AI distinguishes between concepts such as subjective trust, behavioural reliance and task performance. These measures do not always change together. A person may rely on a system without fully trusting it, or express confidence without consistently following its recommendations.

The important question is not simply whether an interface produces the intended behaviour. It is whether users understand what is happening, feel comfortable with it and remain willing to use the system over time.

Familiarity Is Part of Usability

Users gradually develop spatial memory when they interact with an interface. They learn where navigation controls, filters, actions and content are likely to appear.

This knowledge reduces the amount of active attention required to use the product. Instead of repeatedly searching the screen, users can rely on learned locations and familiar interaction patterns.

When an adaptive interface frequently moves controls, reorganises navigation or changes the position of important actions, it interrupts this learning process. Even when each new arrangement has been optimised for a specific situation, the overall experience may become less predictable.

Research and established UX guidance show that spatial consistency helps users remember where interface elements are located. Moving controls can force people to search for functions they previously accessed automatically.

This creates a hidden cost that may not appear in basic analytics. A user might still complete the task, but with more hesitation, greater mental effort and less confidence.

Consistency should not prevent an interface from improving. However, change needs to respect what users have already learned.

Pattern Recognition and Violated Expectations

People do not approach every interface as a completely new environment. They bring expectations formed by years of using websites, applications and operating systems.

They expect certain elements to behave in familiar ways:

• Navigation should remain relatively stable
• Buttons should look interactive
• Headings should indicate hierarchy
• Related information should be visually grouped
• Primary actions should be easy to distinguish
• Similar components should behave consistently

These conventions are not simply decorative traditions. They reduce the amount of interpretation required to understand an interface.

AI-generated layouts can reproduce familiar visual styles while still missing the deeper logic behind them. A system may create a page that looks polished in isolation but contains inconsistent spacing, unclear hierarchy or unusual placement of controls.

Visual design principles such as scale, hierarchy, balance, contrast, and grouping help people understand relationships among elements. When these relationships are weak or inconsistent, a layout may feel wrong before the user can clearly explain why.

This is one reason technically valid AI-generated interfaces can feel less coherent than human-designed ones. Individual components may appear correct, but the relationships among them lack a clear and consistent intent.

Personalisation Can Create Instability

Personalisation can make interfaces more relevant by prioritising content or actions according to user behaviour. However, personalisation becomes problematic when it changes the environment's structure too frequently.

There is an important difference between adapting content and adapting interaction patterns.

Changing the articles, products or recommendations displayed within a stable structure is usually easier to understand. Moving navigation, changing the location of controls or altering the meaning of familiar components can be far more disruptive.

An interface that constantly adapts may become a moving target. Users cannot confidently develop habits because the system may behave differently the next time they return.

This does not mean that adaptive interfaces are inherently poor. It means that personalisation should have boundaries. Core navigation, essential controls and critical workflows should remain stable unless there is a strong reason to change them.

The most useful personalisation often occurs inside a consistent framework rather than through continuous restructuring of the interface.

The Aesthetics of Uniformity

AI-generated interfaces often converge around common patterns because they are trained on large collections of existing designs. This can produce layouts that are clean, usable and technically competent.

However, it can also produce a sense of sameness.

Repeated card layouts, predictable gradients, oversized headings, rounded containers and familiar dashboard structures may work individually but fail to communicate a distinctive identity. The result is not necessarily bad design. It is designed with limited characters.

Human designers frequently introduce variation that is difficult to justify through performance metrics alone. An unexpected shift in scale, a distinctive typographic treatment or a carefully placed area of empty space can create rhythm, emphasis and personality.

These decisions are not random imperfections. When used carefully, they help communicate tone, brand and intention.

The problem is therefore not that AI makes interfaces too perfect. AI-generated work can contain many inconsistencies. The deeper issue is that it often reproduces surface-level conventions without understanding why a particular visual decision is appropriate for a specific product, audience or context.

More Content Is Not Always Better

Optimisation systems may treat unused space as an opportunity to display additional content, recommendations or calls to action.

From a purely numerical perspective, adding more options can create more opportunities for interaction. From a human perspective, however, every additional element competes for attention.

White space supports readability, grouping and visual hierarchy. It helps users identify which elements belong together and which actions are most important. It also gives the interface a sense of pace.

When the screen is fully filled, the user must work harder to distinguish important information from secondary content. The interface may generate more interaction while becoming more mentally demanding.

Good interface design is therefore not about maximising the number of visible opportunities. It is about deciding what deserves attention and what can be left out.

Control and Agency

One of the most important challenges in AI-driven interfaces is the user’s sense of control.

A system may correctly predict what a person wants, but still create discomfort if it does not explain its behaviour or allow the user to change it.

For example, users may become frustrated when:

• Recommendations cannot be corrected
• Personalised settings cannot be disabled
• Interface changes happen without notice
• Automated actions are difficult to reverse
• The system does not explain why something was prioritised
• Users cannot return to a familiar arrangement

Research on algorithmic aversion shows that people may reject algorithmic systems even when they perform well. A 2023 CHI study examined whether different forms of user control could reduce this resistance. It found that the type of control offered to users can influence their willingness to accept algorithmic systems, although simply adding more choices does not automatically create trust.

This suggests that meaningful agency matters more than superficial customisation.

A useful control helps users understand, influence or correct the system. A collection of unexplained settings may create additional complexity without increasing confidence.

Predictability Builds Trust

Trust in an interface is partly built through repeated, understandable behaviour.

When users perform the same action, they expect a reasonably consistent result. When the system changes, they need enough feedback to understand what happened and why.

AI systems can weaken this predictability because their outputs may depend on behavioural data, model updates, changing context or probabilities that are not visible to the user.

From the system’s perspective, the behaviour may be logical. From the user’s perspective, it may appear arbitrary.

Explanations can help, but transparency alone is not a complete solution. Research into human-centred AI explanations shows that explanations may influence reported trust and behavioural reliance in different ways. Their effectiveness also depends on the context, the decision and what the user is trying to achieve.

Designers should therefore avoid assuming that adding an explanation automatically makes an AI system trustworthy. The system must also behave consistently, communicate uncertainty and give users practical ways to respond.

Why AI-Generated Interfaces Can Feel Impersonal

People often describe AI-generated products as sterile or impersonal. This reaction is not necessarily caused by the absence of a human designer. Users usually do not know exactly how an interface was produced.

Instead, the feeling may come from a lack of contextual sensitivity.

A generic interface can appear polished while failing to reflect:

• the seriousness or emotional weight of the task
• the language of the intended audience
• the organisation’s identity
• the user’s level of experience
• accessibility requirements
• cultural expectations
• The consequences of making an error

For example, a playful interaction style may be suitable for an entertainment product but inappropriate for a medical, financial or government service. An interface generated from common patterns may not recognise these differences unless the design process includes clear contextual constraints.

Human-centred design requires more than assembling usable components. It requires understanding what the interaction means to the person using it.

Towards Psychologically Informed AI Design

The solution is not to remove AI from interface design. AI can help teams explore alternatives, analyse complex patterns, identify usability problems and produce variations more quickly.

The challenge is to place this capability inside a human-centred framework.

A psychologically informed approach should include several principles.

Optimise for multiple outcomes: Products should not optimise for a single metric such as engagement or conversion. Evaluation should also consider comprehension, perceived control, confidence, cognitive effort, accessibility and long-term satisfaction.

Preserve structural consistency: Personalisation should usually change content before it changes core interaction patterns. Navigation, essential controls and critical workflows should remain stable enough for users to learn them.

Make the adaptation visible: When an interface changes because of personalisation or AI, the user should be able to understand that a change has occurred and, where appropriate, why.

Provide meaningful control: Users should be able to correct recommendations, undo automated actions, manage personalisation and return to predictable defaults.

Design for trust calibration: The goal should not be to make users trust AI as much as possible. It should be to help them understand when the system is reliable, when it may be uncertain and when human judgement is required.

Keep human review in the process: AI-generated layouts should be reviewed for context, accessibility, clarity, emotional tone and consistency. A design that satisfies a technical prompt is not necessarily ready for real users.

Test the experience, not only the outcome: Behavioural analytics should be combined with interviews, usability testing, satisfaction measures and qualitative feedback. Numbers can identify what happened, but users are often needed to explain why.

The Future Is Collaborative

AI is unlikely to remove the need for interface designers. It is more likely to change where design effort is concentrated.

Generating a layout may become faster. Evaluating whether that layout is appropriate, coherent, accessible and trustworthy will remain a human-centred responsibility.

The strongest interfaces will not simply be those that maximise clicks or minimise task time. They will balance performance with familiarity, clarity, agency and emotional appropriateness.

An interface can be efficient without feeling humane. It can be personalised without feeling predictable. It can be visually polished without expressing a clear intention.

The role of design is to resolve these tensions.

AI can help identify patterns that humans might miss. Human designers must ensure that those patterns are translated into experiences people can understand, learn and trust.

References

1. Nielsen Norman Group. “Spatial Memory: Why It Matters for UX Design.”
   https://www.nngroup.com/articles/spatial-memory/
2. Nielsen Norman Group. “Top 10 Application-Design Mistakes.”
   https://www.nngroup.com/articles/top-10-application-design-mistakes/
3. Cheng, Lingwei, and Alexandra Chouldechova. “Overcoming Algorithm Aversion: A Comparison between Process and Outcome Control.” Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems.
   https://dl.acm.org/doi/10.1145/3544548.3581253
4. Scharowski, Nicolas, et al. “Exploring the Effects of Human-Centered AI Explanations on Trust and Reliance.” Frontiers in Computer Science, 2023.
   https://www.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2023.1151150/full
5. Hoffman, Robert R., Shane T. Mueller, Gary Klein, and Jordan Litman. “Measures for Explainable AI: Explanation Goodness, User Satisfaction, Mental Models, Curiosity, Trust, and Human-AI Performance.” Frontiers in Computer Science, 2023.
   https://www.frontiersin.org/journals/computer-science/articles/10.3389/fcomp.2023.1096257/full
6. Nielsen Norman Group. “5 Principles of Visual Design in UX.”
   https://www.nngroup.com/articles/principles-visual-design/
7. Nielsen Norman Group. “Autonomy, Relatedness, and Competence in UX Design.”
   https://www.nngroup.com/articles/autonomy-relatedness-competence/