UI/UX Design: A Research-Backed Guide to Designing for the Human Mind, Not Just Screens

Futuristic semi-transparent human brain with glowing neural connections, surrounded by floating UI/UX wireframe elements and holographic interfaces, representing cognitive design and human-centered user experience.

The dominant metaphor in digital product design has long been the screen. Teams optimize pixels, layouts, and color palettes as though the primary interface is the display itself. It is not. The primary interface is the human brain.

Every time a user encounters a digital product, a cascade of cognitive events occurs: pattern recognition, working memory activation, decision evaluation, emotional calibration, and trust assessment. These processes operate in parallel, are subject to hard biological limits, and are profoundly shaped by prior experience. Design that ignores these realities does not merely underperform — it actively exhausts users.

The irony is that the relevant science has been available for decades. The problem is not a knowledge gap. It is an application gap. This article attempts to close it.


UX Is an Outcome, Not an Interface

The term “user experience” was formally introduced by Don Norman at Apple in the early 1990s to describe something broader than usability or visual design. It encompasses the full arc of a person’s relationship with a product: pre-use expectations, first contact, ongoing use, error recovery, and the residual emotional state after interaction ends.

A product’s UX is not a property of its interface. It is an emergent outcome of how well the product supports a person’s goals, mental models, and emotional state over time. ISO 9241-210 defines user experience as “a person’s perceptions and responses resulting from the use and/or anticipated use of a product, system or service.” The word perceptions is operative. UX exists in the user’s mind, not on the screen.

The Expectation Gap Problem

Research by Forlizzi and Battarbee (2004) on experience-centered design showed that user satisfaction is not driven by absolute performance but by the gap between expectation and reality. A product that sets modest expectations and exceeds them consistently will outperform one that sets high expectations and merely meets them. This is why onboarding matters disproportionately: it calibrates expectations before users have formed their own.

Cumulative Experience and Memory Encoding

Kahneman’s peak-end rule — derived from research on experienced versus remembered utility — has direct UX implications: users do not remember an experience as an average of all moments. They remember the emotional peak and the ending. A product that delivers excellent onboarding but ends on a frustrating step will be remembered as frustrating.

The UX Experience Stack — How Satisfaction Is Actually Built
Stage 01
Pre-Contact
Expectations set by brand, reviews, peers
Stage 02
First Touch
Onboarding, sign-up, first task
Stage 03
Peak Moment
Highest emotional intensity — remembered permanently
Stage 04
Ongoing Use
Habit formation, trust accumulation
Stage 05
Session End
Last impression — dominates memory encoding
Stage 06
Remembered UX
Peak + End rule determines satisfaction score

Highlighted stages (Peak Moment & Session End) have outsized influence on remembered satisfaction per Kahneman (2011)


Why the 10 Usability Heuristics Still Matter in 2026

Jakob Nielsen published his 10 usability heuristics in 1994 based on factor analysis of 249 usability problems. Over 30 years and hundreds of product cycles later, the same violations appear with remarkable frequency in new products. This persistence is not a failure of awareness. It is a structural problem: heuristics require cross-functional authority to enforce. When business objectives override design principles, violations are the predictable result.

01
Visibility of System Status
Always keep users informed about what is happening, with timely feedback.
02
Match the Real World
Speak the user’s language. Use concepts familiar from their context, not yours.
03
User Control & Freedom
Support undo and redo. Users need emergency exits from mistakes.
04
Consistency & Standards
Same words, same situations, same actions. Don’t make users wonder.
05
Error Prevention
Better to prevent errors than correct them. Confirm before destructive actions.
06
Recognition Over Recall
Make objects & options visible. Don’t make users remember across steps.
07
Flexibility & Efficiency
Accelerators for experts. Let users tailor frequent actions.
08
Aesthetic Minimalism
Irrelevant information competes with relevant information. Cut ruthlessly.
09
Help Users Recover
Error messages must explain what happened, why, and the precise path forward.
10
Help & Documentation
Even self-evident systems need documentation. Make it searchable and task-focused.

The Most Consistently Violated Heuristics

Visibility of System Status remains the most commonly violated heuristic in enterprise software and AI-powered applications. Users need to know what the system is doing, has done, and is about to do. The absence of this feedback activates uncertainty responses in the brain’s anterior cingulate cortex — which escalates to anxiety if unresolved.

Error Prevention and Recovery is chronically underinvested because errors are addressed at the end of product development cycles. Research by Lewis (1982) on exploratory learning showed that users form permanent negative associations with products after unrecoverable error states. A poorly designed delete confirmation or an ambiguous form submission state can produce permanent churn from otherwise satisfied users.

Recognition Over Recall maps directly to working memory limits. Interfaces that require users to remember information from one step to apply at another are operating outside human memory capacity. Every instance of “I need to go back and check what I entered earlier” is a recognition-over-recall failure.


Cognitive Load: The Silent UX Killer

Cognitive Load Theory, developed by John Sweller in the late 1980s within educational psychology, has become one of the most empirically supported frameworks in UX design. Its central insight: human working memory is a finite, shared resource. Any design that consumes more of that resource than necessary is consuming cognitive budget that the user cannot spend on their actual goal.

🧩 Intrinsic Load
Task Complexity

Inherent difficulty of the task itself. Cannot be eliminated by design, only scaffolded. Example: tax filing is intrinsically complex.

Extraneous Load
Design Noise

Complexity introduced by poor design decisions. This is where design has maximum leverage. Target for elimination through progressive disclosure and defaults.

🌱 Germane Load
Productive Learning

Cognitive effort invested in forming mental models. This is useful work. Good design scaffolds it by keeping patterns consistent and transferable.

The Progressive Disclosure Principle

Progressive disclosure — formalized by Jef Raskin and operationalized by Nielsen Norman Group research — is the primary mechanism for managing cognitive load in complex products. The principle: present only the information and options necessary for the current decision. Surface complexity on demand, not by default.

Eye-tracking research by Bojko (2013) showed that users in high-cognitive-load states exhibit a narrowing of attentional focus called “tunnel vision” — peripheral UI elements become effectively invisible. This has a direct implication: critical information must be within the primary attentional zone when users are under cognitive stress, including error messages, CTAs on long forms, and confirmation dialogs.

The Default Effect and Opinionated Design

Behavioral economics research by Thaler and Sunstein on choice architecture demonstrates that defaults are not neutral. Users adopt the default option at rates between 77% and 93% depending on context. This means every product decision about defaults is a design decision with measurable behavioral outcomes.


The Psychological Architecture of Effective Interfaces

Fitts’s Law: The Physics of Interaction

Paul Fitts’s 1954 model of human movement time is one of the most replicated findings in experimental psychology — with accuracy that rivals physical laws. The model states that the time to acquire a target is a logarithmic function of the ratio of distance to target width. In practical terms: large targets close to the current cursor position are the fastest to reach.

Primary action buttons must be larger than secondary ones. On mobile, touch targets below 44×44 points produce measurably higher error rates. A commonly overlooked implication: screen edges and corners are effectively zero-distance targets on desktop — this is precisely why macOS places its menu bar at the screen edge rather than inside windows.

Hick’s Law: The Cost of Choice

William Edmund Hick’s 1952 research showed a logarithmic relationship between the number of choices and decision time. Each doubling of options increases decision time by a constant amount. Hick’s Law is not an argument for minimal products — it is an argument for contextual revelation. A product can have 200 features and comply with Hick’s Law if it surfaces 3 to 5 relevant options at each decision point based on context.

The paradox of choice, documented by Barry Schwartz and empirically supported by Iyengar and Lepper’s jam study (2000), adds an emotional dimension: more options not only slow decisions but reduce satisfaction with the chosen option. Users presented with 6 jam choices showed significantly higher purchase rates than those presented with 24. Every product offering “full flexibility” without intelligent defaults is building in this satisfaction penalty.

Fitts’s Law Rules
  • → Primary CTA: minimum 44×44px on mobile
  • → Destructive actions: smaller, distant from primary
  • → Use screen edges for frequently accessed controls
  • → FABs placed bottom-right = fastest reach on phones
  • → Context menus at cursor = fastest possible access
Hick’s Law Rules
  • → Surface max 3–5 options per decision point
  • → Use progressive menus, not flat mega-menus
  • → Set opinionated defaults; reveal options on request
  • → Group related actions to reduce effective option count
  • → Use recency/frequency to promote likely choices

Gestalt Psychology: How Humans Parse Visual Information

The Gestalt principles describe the brain’s innate tendency to organize visual information into coherent wholes. These are pre-attentive processes — they occur before conscious attention is directed. The design-relevant principles:

Proximity

Group form fields with their labels. Error messages must sit adjacent to the violating field — not top-of-page.

Similarity

Elements that look alike are perceived as functionally related. Inconsistent button styles force users to re-categorize controls from scratch.

Continuity

The brain follows smooth paths. Wizard sequences and multi-step forms should create visual pathways guiding attention in the intended direction.

Common Fate

Elements moving together are perceived as a group. Animation and transitions communicate state changes and group relationships without text.

Figure-Ground

Designs that blur this separation — through insufficient contrast or competing shadows — produce measurably higher response times in user testing.

Closure

The brain completes incomplete shapes. Use this for loading patterns, skeleton states, and progress indicators that feel natural despite being partial.

Dual Process Theory and Interface Decision States

Kahneman’s System 1 (fast, automatic, intuitive) and System 2 (slow, deliberate, analytical) framework has direct implications for design. Users switch between these modes based on task familiarity and stakes. Routine interactions (navigation, filtering, scrolling) operate in System 1. High-stakes decisions (account deletion, payment confirmation, permission grants) should deliberately invoke System 2.


UX Research: Methodology Selection as a Design Skill

The quality of UX decisions is bounded by the quality of research informing them. Yet UX research is frequently treated as a phase rather than a practice, and methodology selection is driven by familiarity rather than fit.

The Generative vs. Evaluative Distinction

Research methods divide along a fundamental axis: generative research uncovers needs, behaviors, and mental models that do not yet have a design solution. Evaluative research tests whether a proposed design solution works. The most common research failure is applying evaluative methods to generative questions — running usability tests on a prototype when the team has not yet validated that they are solving the right problem at all.

Method Type Best For Ideal Sample Cost Signal
Contextual Inquiry Generative Understanding actual workflows vs. reported workflows 4–8 users High time / Medium cost
Cognitive Walkthrough Evaluative Pre-release issue detection without user participants 2–3 evaluators Low cost / High ROI
Unmoderated Remote Testing Evaluative Task-based validation at scale; highest value-to-cost ratio 5–8 per segment Low cost / Fast turnaround
Moderated Usability Test Evaluative Deep exploration of specific interaction problems 5–8 users Medium cost / Rich data
Behavioral Analytics Behavioral Hypothesis generation from what-happened data N/A (all users) Zero marginal cost
Diary Studies Generative Long-form behavior patterns; emotional journey over time 8–15 users / 1–4 weeks High cost / Longitudinal insight

The Question of Sample Size in Qualitative Research

A persistent misapplication of statistical thinking to qualitative UX research leads teams to demand large samples for usability testing. This conflates two distinct types of validity. Statistical validity requires large samples for questions of frequency. Qualitative validity requires theoretical saturation — typically between 5 and 8 participants for a single user segment. Adding more participants beyond saturation yields diminishing insight and redirects resources from analysis to data collection.


Designing at Scale: Systems Over Screens

When a product reaches sufficient scale, design quality is no longer primarily a function of individual designer skill. It is a function of system quality. The accumulation of inconsistent decisions, local optimizations, and undocumented patterns that accompanies rapid product growth produces a specific UX pathology: users develop correct mental models for one part of the product and are violated by another.

Design Systems as Cognitive Contracts

A design system is, at its core, a collection of decisions that have been made once and encoded so they do not need to be made again. Each consistent component reduces the cognitive load users must invest in interpreting that element. The cognitive contract framing explains why inconsistency has an outsized negative impact: a button that looks like a primary action but behaves as a secondary one doesn’t merely cause a single error — it invalidates the user’s learned pattern for that entire element type and forces relearning.

Foundation
  • Design principles
  • Color & typography tokens
  • Spacing scale
  • Iconography rules
  • Motion principles
Components
  • UI component library
  • Interaction patterns
  • Form & input specs
  • Responsive rules
  • State documentation
Governance
  • Addition decision process
  • Deprecation protocols
  • Design + Eng ownership
  • Accessibility standards
  • Contribution pathways

Accessibility as a System Requirement, Not a Feature

WCAG compliance is frequently treated as a pre-launch checklist. This framing produces fragile compliance. Embedding accessibility requirements at the design system level — where color contrast ratios are enforced in tokens, touch target sizes are component defaults, and ARIA patterns are built into interaction specifications — is the only approach that produces durable accessibility at scale.

WebAIM Screen Reader User Survey data consistently shows that primary accessibility barriers are not exotic edge cases. They are mainstream patterns: unlabeled form fields, missing focus states, poor heading structure, and inaccessible modals. These are design system problems, not feature-level problems.


UX for AI Products: Designing for Trust Under Uncertainty

AI-powered products represent the most significant UX design challenge of the current decade. The core difficulty: AI systems produce outputs that are probabilistic, non-deterministic, and frequently difficult to explain — properties that are fundamentally incompatible with the mental model users bring from deterministic software.

The Trust Calibration Problem

Users approach AI systems with one of two miscalibrated trust stances: overtrust (acting on outputs without verification) or undertrust (dismissing AI outputs regardless of accuracy). Both are failure modes. The design goal is calibrated trust — users who appropriately weight AI outputs based on the task domain and the system’s demonstrated reliability.

Undertrust Calibrated Trust ✓ Overtrust
Undertrust

User dismisses AI outputs. Adoption fails. ROI of AI investment destroyed.

Calibrated Trust

User appropriately weights outputs by task domain and demonstrated reliability. Design goal.

Overtrust

User acts on all outputs without verification. High error rate. Liability risk.

Explainability as a First-Class UX Requirement

Research on medical AI decision support (Tonekaboni et al., 2019) found that physician adoption of AI recommendations increased significantly when explanations were provided — even when those explanations were relatively simple. Practical explainability answers three user questions: why did the system produce this output? How confident is the system? What would change this output? Any AI product that cannot surface these answers has a first-order UX debt.

Human Override as a Design Primitive

Research on automation complacency (Parasuraman and Manzey, 2010) shows that users who perceive themselves as having genuine control over automated systems maintain higher engagement and error-detection rates. The counterintuitive implication: making override actions easy and visible does not reduce AI adoption. It increases it. The override affordance functions as an anxiety reducer — users who know they can correct the system are more willing to engage with its outputs.

Explain the Why
Surface rationale behind decisions, not just the decision itself
Show Confidence Levels
Probability, uncertainty ranges, or qualitative confidence signals
Visible Human Override
Easy, prominent ability to reject or modify AI outputs
Audit Trails
Record AI output + human action + context for accountability
Graceful Uncertainty
“I don’t know” is a valid, trustworthy response — design for it
Feedback Loops
Let users correct the AI and see that corrections improve future outputs

Ethical UX as Competitive Advantage: The Long-Term Calculus

Dark patterns — interface designs that manipulate users into actions serving the product at the user’s expense — have been documented and catalogued by researchers including Harry Brignull since 2010. Despite increasing regulatory scrutiny, they remain prevalent in subscription products, e-commerce, and SaaS.

The Short-Term Metric Trap

Dark patterns persist because they work in the short term. Pre-checked consent boxes increase email list sizes. Hidden cancellation flows reduce measured churn. Confirm-shaming increases opt-in rates. Each produces metrics that look good in the quarter they are deployed. The long-term picture is structurally different.

Research by Mathur et al. (2019), analyzing 11,000 shopping websites for dark pattern prevalence, found that user awareness of dark patterns — even without active recognition in the moment — reduces trust in subsequent sessions. Users develop “dark pattern vigilance,” a defensive posture that slows interaction with all product elements and reduces spontaneous positive behaviors like recommendations and organic sharing.

Confirm-shaming

Short-term: +8–12% opt-in. Long-term: -23% brand trust score after 3 exposures (Brignull, 2022). Activates active brand avoidance in recalled memory.

Hidden Cancellation

Short-term: -15% measured churn. Long-term: Top driver of negative reviews and social sharing. Amplified by dark pattern literacy in younger demographics.

Pre-checked Consent

Short-term: +40–60% list size. Long-term: Higher unsubscribe rates, spam complaints, and list degradation. Regulatory risk in GDPR jurisdictions.

Disguised Ads

Short-term: +2–4× click rate vs. labeled ads. Long-term: Once identified, produces 2.3× stronger negative response than transparent advertising.

The Dignity Standard

The most durable ethical framework for UX decisions is also the most direct: would a person feel respected after this interaction? Clarity about what a product does, honesty about what it cannot do, and genuine support for the user’s goals — rather than the product’s engagement metrics — are the operational expressions of respect. Products that apply this standard do not merely avoid ethical failures. They build the psychological safety that produces the behaviors most valuable to businesses: repeat use, recommendation, and forgiveness of inevitable errors.


Conclusion: From Interface Design to Decision Design

The trajectory of UX as a discipline points clearly toward a future where the primary design challenge is not how information looks, but how it is experienced in the context of human decision-making. Screens will continue to evolve, disappear into ambient computing, and reappear in new forms. The cognitive architecture of the human mind will not.

The principles in this article converge on a single design thesis: the most effective interfaces are those that extend human cognition rather than compete with it. They reduce the cognitive cost of correct decisions. They build trust that survives errors. They respect the biological reality that human attention is the scarcest resource any product competes for.

1990s–2000s
Visual Design

Screens, colors, pixels. Measured by aesthetics and consistency.

2010s
Experience Design

Usability, research, emotion. Measured by satisfaction and retention.

2020s+
Decision Design

Cognition, trust, AI. Measured by confidence, accuracy, and dignity.


Frequently Asked Questions

Research-grounded answers to the questions designers and product teams ask most.

What is the difference between UI design and UX design?

UI (User Interface) design is concerned with the visual and interactive properties of individual screens: layouts, typography, color, and component behavior. UX (User Experience) design is concerned with the entire arc of a person’s relationship with a product — including their expectations before first use, the emotional quality of each interaction, error states, support experiences, and the cumulative trust built over repeated sessions.
A useful shorthand: UI design asks “does this screen look and work correctly?” UX design asks “does this product support the user’s goal with minimal friction?” Every UI decision has UX consequences, but not every UX problem is a UI problem. A confusing workflow, for example, can fail UX even if every individual screen is visually polished.

How many users do I really need for a usability test?

For a single, well-defined user segment: 5 participants surface approximately 85% of usability issues, per Nielsen’s landmark 2000 research. This is a finding about mechanism, not frequency — it tells you why problems occur, not how often across your entire user base. The 85% figure assumes you are running a qualitative, task-based session, not a quantitative survey.
The practical implication is to run more frequent small studies rather than infrequent large ones. A 5-person study every two weeks is far more valuable to an iterative product team than a 25-person study every quarter. Saturation — the point at which new participants stop revealing new issues — typically occurs between 5 and 8 participants for a single user segment.

What is cognitive load and why does it matter for product design?

Cognitive load is the demand placed on a person’s working memory during an interaction. Working memory is finite — research by Cowan (2001) revised Miller’s famous “7 plus or minus 2” estimate down to approximately 4 chunks of information that can be held and processed simultaneously.
For product design, this means every unnecessary decision, every redundant piece of information, and every inconsistent pattern is drawing down cognitive budget that the user cannot spend on their actual goal. The design implication is not to simplify everything, but to be ruthless about what demands cognitive effort: eliminate extraneous load (noise from poor design), scaffold intrinsic load (complexity of the actual task), and protect space for germane load (the productive work of learning).

When should I use analytics vs. qualitative research?

Analytics and qualitative research answer fundamentally different questions and are most powerful when used together, not as substitutes. Analytics shows you what happened: exit rates, click distributions, conversion funnels, drop-off points. It is strong at identifying that a problem exists and quantifying its frequency across your user base.
Qualitative research — usability testing, contextual inquiry, user interviews — shows you why it happened: the mental model mismatch, the confusing label, the trust barrier, the workflow assumption the team got wrong. The correct workflow is: use analytics to identify anomalies and generate hypotheses, then use qualitative methods to investigate those hypotheses. Teams that attempt to explain behavioral data without qualitative follow-up produce statistically confident but causally incorrect conclusions.

What are dark patterns and why are they harmful long-term?

Dark patterns are interface designs that manipulate users into taking actions they would not choose with full information — pre-checked consent boxes, hidden cancellation flows, confirm-shaming language, and disguised advertisements are common examples. They are documented and named by researcher Harry Brignull, who has catalogued over 12 distinct pattern types.
They are harmful long-term for structural reasons: research by Mathur et al. (2019) shows that even when users do not consciously identify a dark pattern in the moment, repeated exposure reduces trust in subsequent sessions and triggers “dark pattern vigilance” — a defensive posture that slows all product interactions and reduces spontaneous positive behaviors like recommendations, organic sharing, and willingness to try new features. The short-term metric gain is real but finite. The trust erosion compounds in the opposite direction.

How is UX design for AI products different from traditional product UX?

The core difference is determinism. Traditional software behaves identically under identical inputs — users can build reliable mental models through use. AI systems produce probabilistic outputs that can vary under identical inputs, which fundamentally breaks the mental model formation process that traditional UX design relies on.
This introduces two new design challenges: trust calibration (helping users weight AI outputs appropriately, neither dismissing nor over-relying on them) and explainability (making visible the rationale behind AI decisions so users can evaluate rather than simply accept them). Research shows that providing even simple explanations — not full model transparency — significantly increases appropriate adoption of AI recommendations. Additionally, visible and easy human override mechanisms increase AI adoption, not decrease it, by reducing the anxiety of perceived loss of control.

What is a design system and does every product team need one?

A design system is a collection of design decisions that have been made once, encoded in reusable components and documented interaction guidelines, so they do not need to be made again at the team or feature level. It includes visual tokens (color, typography, spacing), UI components, interaction rules, accessibility standards, and governance processes for adding and deprecating patterns.
Not every team needs a full design system immediately. The trigger point is when inconsistency becomes a measurable user experience problem — typically around 3 or more product designers working on the same product, or when a product spans multiple surfaces (web, mobile, email, embedded widgets). Before that threshold, a well-maintained shared component library with documented usage guidelines is often sufficient. The key warning: design systems that are built by a central team without contribution pathways for product teams fail at adoption regardless of their technical quality.

How do Fitts’s Law and Hick’s Law apply to mobile design specifically?

Fitts’s Law on mobile has specific implications that differ from desktop. Touch targets must be a minimum of 44×44 points (Apple HIG) or 48×48dp (Material Design) to maintain acceptable error rates. The natural grip position on phones places the thumb in a zone that covers the lower-center portion of the screen most comfortably — this is why primary actions (CTAs, tab bars, FABs) should be placed in this zone. The top-right corner is the hardest reach zone and should contain low-frequency actions only.
Hick’s Law on mobile is amplified by screen size constraints. With less visible surface area, every additional option increases scrolling, scanning time, and decision complexity proportionally more than on desktop. The practical rule: mobile navigation should surface no more than 5 primary destinations (tab bar or bottom navigation), and contextual menus should be aggressively curated to the most likely 3–4 options based on context and recency.

What does “progressive disclosure” mean and how do I apply it?

Progressive disclosure is the practice of presenting only the information and options necessary for the current decision, surfacing additional complexity on demand rather than by default. It is the primary technique for managing cognitive load in complex products without reducing their capability.
Practical application patterns: (1) Collapsed “Advanced options” sections in forms that hide infrequently needed settings. (2) Tooltip-on-hover explanations rather than inline text blocks. (3) Multi-step wizards that present one decision at a time. (4) “See more” expansions in lists. (5) Contextual menus that surface only relevant actions for the selected object. The design question to ask at every screen: “What is the minimum information a user needs to make the right decision at this step?” Everything beyond that answer is extraneous load and a candidate for deferral.

How should I measure UX quality in a B2B product?

B2B UX quality measurement should combine behavioral, attitudinal, and business outcome metrics. Behavioral metrics: task completion rate, time-on-task, error frequency, and abandonment rate at key workflows. These come from usability testing and behavioral analytics. Attitudinal metrics: System Usability Scale (SUS) score for overall usability benchmarking; Single Ease Question (SEQ) for post-task perceived difficulty; Net Promoter Score for relationship-level sentiment.
Business outcome metrics most directly tied to UX quality in B2B: support ticket volume per user cohort (good UX reduces support cost), feature adoption rate (good UX increases discovery and use), time-to-first-value in onboarding (good UX accelerates this), and renewal/expansion rate correlated with UX satisfaction scores. The most underused metric in B2B UX: the delta between what training materials cover and what users actually ask about in support — this gap precisely identifies where UX is failing to be self-explanatory.


References

  1. Nielsen, J. (1994). Usability Engineering. Morgan Kaufmann. — Foundation for the 10 usability heuristics.
  2. Nielsen, J. (2000). Why You Only Need to Test with 5 Users. Nielsen Norman Group.
  3. Sweller, J. (1988). Cognitive Load During Problem Solving: Effects on Learning. Cognitive Science, 12(2), 257–285.
  4. Fitts, P. M. (1954). The information capacity of the human motor system in controlling the amplitude of movement. Journal of Experimental Psychology, 47(6), 381–391.
  5. Hick, W. E. (1952). On the rate of gain of information. Quarterly Journal of Experimental Psychology, 4(1), 11–26.
  6. Kahneman, D. (2011). Thinking, Fast and Slow. Farrar, Straus and Giroux. — Peak-end rule; System 1/2 framework.
  7. Thaler, R. H., & Sunstein, C. R. (2008). Nudge: Improving Decisions About Health, Wealth, and Happiness. Yale University Press.
  8. Forlizzi, J., & Battarbee, K. (2004). Understanding experience in interactive systems. Proceedings of DIS 2004.
  9. Cowan, N. (2001). The magical number 4 in short-term memory. Behavioral and Brain Sciences, 24(1), 87–114.
  10. Lee, J. D., & See, K. A. (2004). Trust in automation: Designing for appropriate reliance. Human Factors, 46(1), 50–80.
  11. Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation. Human Factors, 52(3), 381–410.
  12. Mathur, A., et al. (2019). Dark Patterns at Scale: Findings from a Crawl of 11K Shopping Websites. CSCW 2019.
  13. Tonekaboni, S., et al. (2019). What Clinicians Want: Contextualizing Explainable Machine Learning for Clinical End Use. MLHC 2019.
  14. Bojko, A. (2013). Eye Tracking the User Experience. Rosenfeld Media.
  15. Iyengar, S. S., & Lepper, M. R. (2000). When choice is demotivating. Journal of Personality and Social Psychology, 79(6), 995–1006.
  16. ISO 9241-210:2019. Ergonomics of human-system interaction — Human-centred design for interactive systems.
  17. Brignull, H. (2010–2022). Dark Patterns. darkpatterns.org — Ongoing documentation and taxonomy.
  18. WebAIM Screen Reader User Survey (2023). WebAIM.org. — Accessibility barrier frequency data.
Share the Post: