The Human Mind. Design for the brain people actually have.

Warm-up: what do you already believe?

Before the theory, test your intuitions. Psychology is the one design subject where everyone already has strong opinions, and many popular ones are half right. Answer quickly; you will meet each of these again later, and noticing where you changed your mind is how the learning sticks.

The eye does not photograph your screen. It samples a tiny sharp spot, fills the rest with prediction, and builds meaning from edges and contrast, so your layout is understood mostly by what the brain expects.

Hold your thumb at arm's length. The nail covers roughly the part of the world you see in sharp detail: the fovea, about 1 to 2 degrees of visual angle. On a laptop at normal distance, that is a patch only a few words wide. Everything else is peripheral vision: blurry, poor at color, but very good at motion and at big shapes. Your eyes jump three or four times a second (saccades) and the brain stitches those snapshots into the illusion of a stable, sharp page.

Helmholtz called perception "unconscious inference" in the 19th century, and the psychologist Richard Gregory built on it: what you see is the brain's best hypothesis about what is out there, given the sparse input and everything it has seen before. That is why people "see" a button that is just a colored rectangle, why a hairline under text reads as a link, and why users miss a warning written in the place they never look.

Edges and contrast come first

Early vision is built to find edges: places where brightness changes. Your retina and visual cortex respond to differences much more than to absolute values. That is why the same gray looks lighter on a dark background (simultaneous contrast), and why a border and a change of background tone are equally good at saying "this is a separate thing". It is also why low-contrast interfaces fail: when the luminance difference between text and background is small, the edge-detecting machinery has less signal, reading slows, and the page feels foggy. Contrast ratios and their exact thresholds belong to Accessible & Inclusive and Visual Design; here the point is the mechanism: luminance contrast is the raw material of every edge, shape and group.

Expectation is part of the input

Because perception is inference, prior experience changes what people see. A blue underlined phrase is a link before it is read. A gray rounded rectangle at the bottom of a phone screen is a home indicator, not a control. An "X" in the top-right corner closes things. When you break these expectations, you are not just being original: you are asking the brain to override a confident prediction, which costs time and causes errors. This is the perceptual root of Jakob's law (chapter 18).

In the wild

• iOS 7Apple, 2013
  The flat redesign removed button borders and made many buttons plain tinted text. Apple later added an accessibility setting ("Button Shapes") to put back visible shapes. When you remove edges, some people lose the signal that something is pressable.
• iA WriterInformation Architects, 2010–
  The writing view is almost only text, but the cursor is a strong, saturated blue bar. The one thing you must find instantly has the most contrast. Minimal does not mean low contrast everywhere; spend contrast on what matters.
• Google Search resultsGoogle
  Blue result titles have looked like links for decades. Each time a variation is tested, the conventional pattern carries enormous expectation. A strong convention is free comprehension you inherit.

People do not see twenty elements; they see four groups. Nearness, likeness, shared containers and connecting lines decide which groups, and your layout means whatever those groups say.

The Gestalt psychologists in Germany in the 1910s and 1920s showed that perception organizes before it identifies. Wertheimer's 1923 paper laid out "laws of organization": rules by which the visual system decides what belongs together. A century later these are among the most reliable findings in all of design psychology, because you can demonstrate them on anyone in seconds. Stephen Palmer added two more in the 1990s, common region and uniform connectedness, that turn out to be the strongest of all for interfaces.

Why it matters: grouping is how people answer "what is related to what?" without reading. If a label is closer to the wrong field, people attach it to the wrong field. If two buttons look the same, people assume they do the same kind of thing. The structure your code has (which div contains what) is invisible. Only the perceived grouping exists for the user.

Proximity: spacing is meaning

Proximity is the cheapest grouping tool you have, because it uses empty space instead of ink. The rule of thumb for forms: the gap between a label and its own field should be clearly smaller than the gap between that field and the next label. A 4 px label gap against a 20–24 px gap between fields (roughly 1:5) reads unambiguously. Equal gaps (say 12 px everywhere) make every label float between two fields.

Similarity: same look, same job

Things that share color, shape, size or type style are seen as belonging together, and, crucially, as doing the same kind of thing. That is a promise. If every tag is a gray pill and one gray pill is actually a button that deletes, you have broken the promise. Similarity is also why consistency matters so much: when two buttons that do different things look identical, people assume they are equivalent.

Common region and connectedness: the strong glue

Palmer showed in 1992 that elements inside a shared boundary group together even when proximity says otherwise, and with Irvin Rock (1994) that elements joined by a line group more strongly still. In interfaces these are the card, the panel, the inset grouped list, the line between steps. They are powerful, so they are also noisy: a screen where everything sits in a bordered box has many regions and therefore no hierarchy. Calm interfaces such as iA Writer, Things and Apple's Settings mostly use proximity and a few large regions, not a box around every element.

In the wild

• iOS SettingsApple, inset grouped lists
  Related settings sit in rounded regions separated by generous gaps and small uppercase headers. Common region plus proximity organizes hundreds of settings without a single border line around a row.
• Things 3Cultured Code, 2017
  Headings inside a project group to-dos almost entirely by whitespace and a thin heading rule. Proximity alone can carry structure in a calm product.
• Stripe CheckoutStripe
  Card number, expiry and CVC are joined into one bordered block with internal dividers. Common region tells you "these three fields are one thing: your card".

The visual system completes broken shapes, follows smooth paths, separates a figure from its ground, binds things that move together, and always prefers the simplest reading. Each of these is a lever for making an interface feel obvious.

Chapter 2 was about which things belong together. This chapter is about which shapes, layers and movements the brain constructs. These principles explain why a half-visible card invites a swipe, why a modal over a dimmed page feels like it is "on top", and why a selected batch of files feels like one object when you drag it.

Continuity and alignment

The eye prefers to follow smooth, continuous lines and edges. In layout, that makes alignment a grouping force: elements whose left edges sit on one invisible line are read as a column, and one element that is 3 px off breaks the line and looks like a mistake. It also explains horizontal scrolling rows: when cards continue off the edge of the screen, the eye follows the row and infers there is more.

Closure: what the brain fills in

Given a partial outline, people perceive the complete shape. This is why icons can be drawn with very few strokes, and why the "peek" works: in a horizontal carousel, a card cut off at the right edge is completed by the mind, which also tells it "there is more here, scroll". When every card fits exactly, nothing hints at more, and people may never swipe.

Figure-ground: what is on top

Edgar Rubin's 1915 vase/faces figure shows that the brain must decide which region is the object and which is the background, and it cannot see both at once. Smaller, enclosed, higher-contrast and lower regions tend to become figure. Interfaces use this constantly: a modal on a dimmed page, a popover with a shadow, a selected row with a filled background. If the ground is not pushed back (the page behind a dialog at full brightness), people are unsure which layer they are in, and click the wrong one.

Common fate: things that move together

Elements that move in the same direction at the same time are grouped, overriding even color and proximity. This is why animation can explain structure: when you collapse a section and its children slide up together, the user learns that they belonged to it. When you drag five selected files and they gather into a stack under the cursor, they become one object. Motion design builds on this in Motion & Feel.

Prägnanz: the simplest reading wins

"Prägnanz" (roughly "conciseness") is the Gestalt umbrella law: the brain resolves ambiguity in favor of the simplest, most regular interpretation. Designers use it two ways. First, give people simple forms so they do not have to resolve anything. Second, when your layout has an accidental simple reading (two unrelated panels that happen to align into a single block), people will see that reading, not yours.

In the wild

• App Store "Today" and shelf rowsApple
  Horizontal shelves size items so the next one is visibly cut at the edge. Closure tells you there is more without arrows or "see all".
• macOS FinderApple
  Dragging several selected files shows them gathered under the pointer with a count badge. Common fate turns many items into one draggable object.
• iOS sheetsApple, iOS 13 and later
  A sheet slides up while the page behind scales back slightly and darkens. Figure-ground made physical: you know which layer you are in.

A few visual features (color, size, orientation, motion, and some others) are detected in parallel across the whole screen in a fraction of a second. Anything that differs by one of them pops out. Anything that differs only by a combination must be searched for, one item at a time.

In 1980 Anne Treisman and Garry Gelade published feature integration theory. Their experiments showed two kinds of search. When the target differs from distractors by a single feature (a red item among blue ones), search time stays roughly flat no matter how many distractors there are: the target "pops out". When it differs only by a conjunction (a red circle among red squares and blue circles), search time grows with the number of items, as if attention must visit each one. The features that pop out are called preattentive: they are processed before focused attention, in roughly 200–250 ms.

For design, this is a budget. Pop-out is the most powerful tool you have for guiding the eye, and it only works when one thing differs. Make five things pop and none of them do. Try it yourself below.

The preattentive toolkit

Researchers have listed many preattentive features. The ones you will use in interfaces are:

In the wild

• Apple Mail / iOS MailApple
  Unread messages get a single blue dot and bolder sender name in an otherwise quiet list. One feature, redundantly coded, scanned at a glance.
• Xcode, VS Code guttersApple, Microsoft
  Errors appear as red marks in the scrollbar and gutter, so you can find them in a 3,000-line file by pop-out rather than reading. Preattentive marks turn search into a glance.
• LinearLinear, 2019–
  The issue list is mostly grayscale; color is kept for small status and priority icons. A quiet baseline makes small colored marks readable.

People do not notice changes they are not attending to, even large ones, and they do not see unexpected things while focused on a task. An update that happens silently on screen is, for most users, an update that never happened.

In 1997 Ronald Rensink, Kevin O'Regan and James Clark showed people two versions of a photo alternating with a brief blank screen in between. Large changes (a whole engine missing from an airplane) took many seconds to spot. Without the blank, the change popped out instantly, because the motion signal of the change drew attention. The blank (like a page reload, a blink or a scroll) erases that signal. This is change blindness.

In 1999 Daniel Simons and Christopher Chabris asked people to count basketball passes in a video. About half failed to notice a person in a gorilla suit walking through the scene. This is inattentional blindness: when attention is committed to a task, unexpected things are not seen at all, even in the fovea. Mack and Rock's 1998 book named the effect after years of lab work.

Both effects are robust and replicate well. They demolish the designer's assumption that "it is on the screen, so they saw it".

Where change blindness bites in software

• Full reloads after an action. The page blanks, re-renders, and the one line that changed ("Shipping: ₹ 120") is invisible.
• Silent recalculation. A total updates when a coupon fails, in a corner the user is not looking at.
• Validation errors far from the field, shown at the top of a long form after submit, while attention is at the bottom on the button.
• Toasts and snackbars that appear in a corner while the user is focused on the content, and vanish in 4 seconds.
• Auto-saved settings with no confirmation near the control.

In the wild

• Google DocsGoogle
  Collaborators' cursors and selections appear in color with name labels right where they are editing. Changes made by others are made visible at the place of change.
• Single-page web appsa pattern
  Inline validation that appears next to a field as you leave it replaced many full-page "submit and reload" forms. No reload, no blank, no change blindness.
• Figma multiplayerFigma
  You see other people's cursors moving live, so changes to the canvas have a visible cause. Motion of the cause makes the change noticeable and explainable.

People can consciously attend to one thing at a time. Your interface either supports that one thing or competes with it, and anything that looks like an ad, a promotion or a nag is learned to be invisible.

Jef Raskin, who started the Macintosh project at Apple, put it plainly in The Humane Interface (2000): we have one "locus of attention", the single object or idea we are consciously engaged with. You can walk and talk, but you cannot attend to two conversations. Skills that are fully automatic (typing for a fast typist) do not need the locus, which is why they can run alongside thought. Everything else queues.

Herbert Simon made the economic version of the point in 1971: "a wealth of information creates a poverty of attention". Each element you add to a screen draws a little from the same fixed budget. Kahneman's Attention and Effort (1973) treated attention as a limited capacity that is spent by effortful tasks; when one task is demanding (composing a hard email), people miss things that would be obvious when relaxed.

Banner blindness: attention learns to ignore

In 1998, Jan Panero Benway and David Lane found that people searching a site failed to find information placed in a large, colorful banner at the top, even though it was exactly what they were looking for. Later eye-tracking by Nielsen Norman Group repeatedly showed users skipping anything that looks like an ad: boxes in the right rail, animated panels, and content styled with promotional colors or stock photos. The visual system learns which shapes are noise for the task and stops looking at them.

Banner blindness spreads beyond ads. Any element that repeatedly offers nothing useful becomes invisible: the "What's new" panel, the persistent upgrade bar, the tips sidebar. Worse, when you later put something important in that place or style, it inherits the blindness.

Designing for one locus

• Put feedback where attention already is. Raskin's strongest practical advice: messages belong at the locus, not in a status bar people never read.
• Avoid modes the user must remember. Raskin argued that modes cause errors because the mode indicator is never at the locus; you type expecting one behavior and get another (the Caps Lock problem).
• Remove instead of styling. Each element costs filtering effort even when ignored. Removing a permanent banner frees more attention than making it subtler.

In the wild

• Vim and modal editorsa classic case
  Powerful, but the mode indicator lives in the bottom line, away from the cursor, and new users type commands as text. Raskin's warning about modes in its purest form; experts survive it by making mode-switching automatic.
• SuperhumanSuperhuman, 2017–
  Shows one email at a time in focus, with keyboard commands surfaced in place (a command palette) rather than toolbars. Designed around a single locus.
• iA Writer Focus ModeInformation Architects
  Dims everything except the current sentence or paragraph. The interface literally draws the locus of attention.

People rarely read screens; they scan for words that match their goal. Where they scan depends on how the page is structured, and when the page gives them nothing to grab, they fall back to the worst pattern: the F.

Nielsen's 2006 eye-tracking study of web pages produced the famous F-shaped pattern: two horizontal sweeps across the top, then a vertical scan down the left edge. Nielsen Norman Group's 2017 follow-up, by Kara Pernice, made the important correction that is often lost: the F-pattern is a symptom of poor structure, not a goal. People fall into it when text has no headings, bold keywords or lists to anchor on. The same research described other patterns: layer-cake (eyes jump from heading to heading, reading body text only under relevant ones), spotted (skipping around for a specific word, number or link), and commitment (reading everything, when motivation is high).

What about the Z-pattern?

The Z-pattern (eyes go top-left to top-right, diagonally down, then across the bottom) and its cousin the Gutenberg diagram are layout heuristics for sparse pages with few elements, like a landing page hero or a poster. They are not strong eye-tracking findings in the way the F and layer-cake are. Use them as a composition idea (logo top-left, action bottom-right, a strong diagonal), not as a law. On dense, text-heavy screens, scanning follows content structure, and the Z disappears.

Design for the layer-cake

1. Write headings that carry meaning. "Export your notes as Markdown" beats "Export". A scanner reads only the heading; it must answer "is this for me?".
2. Front-load. The first two words of every heading, list item and link do most of the work, because scanning eyes fixate on the left edge.
3. Chunk body text short. Two to four sentences per paragraph, one idea each.
4. Make the scannable things look scannable: bold key terms sparingly, numbers as digits ("3 days", not "three days"), lists for parallel items.
5. Put the action next to the explanation, not in a different region, so the layer-cake reader finds it in the same chunk.

Writing for scanning is covered deeply in Words and Reading & Writing. Here the lesson is perceptual: the eye goes to anchors, and you decide where the anchors are.

In the wild

• GOV.UKGovernment Digital Service, 2012–
  Short pages, plain descriptive headings, front-loaded link text, numbers as digits. A whole service built for the layer-cake reader.
• Stripe DocsStripe
  Each section starts with a heading that names the task, followed by a short paragraph and a code block. Developers scan for their task, read one chunk, copy the code.
• Apple Settings search resultsApple
  Results show the setting name first and its location path in smaller gray text. Front-loading the distinguishing word supports spotted scanning.

Deep work happens in a fragile state called flow. Every interruption breaks it, and getting back costs far more than the interruption itself. A product that respects attention protects flow and batches everything else.

Mihaly Csikszentmihalyi studied people at their best, painters, chess players, surgeons, rock climbers, and described flow in his 1990 book Flow: The Psychology of Optimal Experience: total absorption in an activity, loss of self-consciousness, distorted sense of time. Flow appears under specific conditions: a clear goal, immediate feedback, and a balance between challenge and skill. Too easy and people get bored; too hard and they get anxious.

The cost of an interruption

Gloria Mark at UC Irvine has studied interruptions in real offices since the early 2000s. Her field work found that people switch activities every few minutes and that interrupted work is often not resumed for a long time; the popular figure of "about 23 minutes to get back on task" comes from her research and interviews. In a 2008 lab study, Mark, Gudith and Klocke found something subtler: interrupted people completed tasks in about the same time, but by working faster, with significantly more stress, frustration and effort. Interruptions do not just cost time; they cost wellbeing.

Notifications are interruptions you design. Each one asks the user to move their locus of attention from their task to yours. The design question is always: is this worth breaking what they are doing right now?

Calm notification design

Mark Weiser and John Seely Brown at Xerox PARC proposed "calm technology" in 1995: technology that informs without demanding focus, moving easily between the periphery and the center of attention. Amber Case turned it into principles in her 2015 book Calm Technology. In practice for notifications:

In the wild

• iOS Focus and Notification SummaryApple, iOS 15, 2021
  Let people silence apps by context and deliver non-urgent notifications in a scheduled summary. The platform now batches for you; design notifications that make sense in a summary.
• Slack Do Not Disturb and scheduled sendSlack
  Senders can schedule messages for working hours and recipients can pause notifications. Both sides of an interruption get control.
• KindleAmazon, 2007–
  The reading screen shows almost nothing but the page; progress and menus appear only on tap. A reading device built for flow: clear goal, no interruptions, immediate page-turn feedback.

People can hold only about four meaningful chunks in mind at once, for a few seconds. Never make them carry information from one screen to another, and when they must handle long strings, chunk them.

In 1956 George Miller published "The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information" in Psychological Review. The paper is witty and careful. It reviewed two kinds of limits: how many categories people can reliably distinguish on one dimension (absolute judgment, around seven), and how many items people can repeat back in order (immediate memory span, also around seven). Miller pointed out that the coincidence might be "only a pernicious, Pythagorean coincidence". His most lasting idea was chunking: memory span is limited in chunks, not in raw bits, so recoding information into bigger meaningful units extends what we can hold.

In 2001 Nelson Cowan reviewed decades of later research in "The magical number 4 in short-term memory" (Behavioral and Brain Sciences). When you stop people from rehearsing and grouping, the true capacity of the focus of attention is about 3 to 5 chunks, typically 4. Seven was the span with rehearsal and chunking helping.

Chunking: making four go further

A chunk is a unit that is meaningful to the person. "9 8 7 6 5 4 3 2 1 0" is ten digits but one chunk ("counting down"). An unfamiliar 10-digit string has ten. Formatting helps people build chunks: grouping digits by 3 or 4, using separators, keeping known structure (country code, area code) visible. This is why phone numbers, credit cards and IBANs are printed in groups.

The example IBAN above is the standard sample number used in documentation, not a real account.

Cognitive load: three kinds

John Sweller's cognitive load theory (from 1988, developed through the 1990s) divides the mental effort of a task into three kinds. Intrinsic load comes from the task itself: splitting a bill among five people with different shares is inherently harder than splitting it equally. Extraneous load comes from how the task is presented: hunting for a field, decoding jargon, remembering a value from a previous screen. Germane load is effort that builds lasting understanding (learning how the tool's model works). Later versions of the theory folded germane load into intrinsic, but the design rule survives: you cannot remove intrinsic load, you must ruthlessly remove extraneous load, so people's limited working memory goes to their actual problem.

In the wild

• iOS and Android OTP autofillApple iOS 12, 2018; Android SMS Retriever
  Codes received by SMS are offered above the keyboard. The best working-memory design is the one that holds nothing.
• Stripe Elements card fieldStripe
  Formats the card number into groups of four as you type and detects the brand. Chunking applied live, plus Postel's law (chapter 18): it accepts what you type and formats it.
• UPI payment screensIndia, NPCI ecosystem
  Good UPI apps show the payee's registered name after you enter a UPI ID, before you pay. Recognition replaces having to remember and verify a string yourself.

Recognizing something you see is far easier than recalling it from nothing. And people operate your product through a mental model, a private simplified theory of how it works, which your interface either teaches correctly or teaches wrong.

Ask someone to name all the menu items in their email app and they will struggle. Show them the menu and they will instantly know which one they want. Recognition needs only a match between what you see and what you know; recall must reconstruct the item from scratch with only a vague cue. This is why graphical interfaces beat command lines for occasional use, and why Nielsen made "recognition rather than recall" one of his ten usability heuristics (1994).

Recognition patterns you can use

• Visible options instead of hidden commands: menus, toolbars, a command palette that lists actions as you type.
• Recent and frequent items: recent files, recent searches, recently used emoji.
• Autocomplete and previews showing actual content (file thumbnails, link previews).
• Keyboard shortcuts shown next to menu items, so recognition slowly trains recall for those who want speed.
• Breadcrumbs and titles that tell people where they are, so they do not have to remember how they got there.

Long-term memory: kinds that matter for design

Endel Tulving distinguished episodic memory (events you lived) from semantic memory (facts you know); researchers also describe procedural memory (skills and habits). For interfaces: semantic memory holds conventions ("trash can deletes"), procedural memory holds motor habits ("Cmd-S saves"), and episodic memory holds the experience ("last time this app lost my draft"). You build on the first, must not break the second (chapter 11), and are judged by the third (chapter 17).

Mental models: the theory in the user's head

The Scottish psychologist Kenneth Craik suggested in 1943 that the mind builds "small-scale models" of reality to predict what will happen. Don Norman applied this to design in The Design of Everyday Things with three models: the designer's model (how you think it works), the system image (what the product shows: UI, words, behavior, docs), and the user's model (what they infer). You never talk to the user directly. Everything they learn comes through the system image, so if the system image is unclear or inconsistent, the user builds a wrong model, and then makes "errors" that are perfectly logical given that model.

Common broken models

In the wild

• Raycast and command palettesRaycast; VS Code; Linear
  Type a few letters and see every matching action with its shortcut. Recognition-based access to a recall-based world; the shortcut labels gradually teach recall.
• DropboxDropbox, 2008
  Made sync feel like "a folder on your computer", matching the model people already had of files and folders. Borrowing an existing mental model is faster than teaching a new one.
• Gmail Undo SendGoogle, Labs 2009, default 2015
  Gmail cannot unsend an email that has left; it delays sending by a few seconds. Users' model ("I can unsend") is wrong, but the system makes it effectively true. Sometimes you change the system to fit the model people already have.

Repeated actions become automatic habits that run without attention, so moving a control breaks people in a way no amount of explanation fixes. Memory also favors the first and last items in a sequence. And unfinished tasks pull at us, though less magically than the internet claims.

Habituation and muscle memory

Raskin argued that every interface eventually becomes habitual, and that this is a good thing: habits free the locus of attention for the real work. A writer who saves with Cmd-S does not think about saving. But habits run on location and sequence, not meaning. If you move "Save" or reorder a tab bar, the habit fires anyway and lands on the wrong thing. People do not just get slower; they make errors, feel stupid, and blame the product.

Habits also defeat confirmation dialogs. When "Are you sure?" appears every time, the click on "Yes" joins the habit, and the dialog stops protecting anything. Raskin's alternative: make actions undoable instead of confirmable.

Serial position: first and last stick

Ebbinghaus noticed it in the 1880s and Bennet Murdock measured it cleanly in 1962: when people try to recall a list, they remember the first few items (primacy, which had more rehearsal and reached long-term memory) and the last few (recency, still in working memory) much better than the middle. The effect is robust. Try it.

Design uses: put the most important items of a list, a tab bar or an onboarding sequence at the start and end. In a short tab bar, the first and last positions are both the most memorable and the easiest to reach. In a long settings list, the middle is where rarely-needed items belong. In a pitch or a landing page, the opening and the final summary carry the most weight.

The Zeigarnik effect, honestly

In 1927 Bluma Zeigarnik, working in Kurt Lewin's lab in Berlin, reported that people remembered interrupted tasks better than completed ones. Her colleague Maria Ovsiankina (1928) found a related effect: people tend to spontaneously resume interrupted tasks. Product writing often cites "Zeigarnik" to justify progress bars, incomplete profile meters and cliffhangers.

The evidence is weaker than that story suggests. Replications over the decades were mixed, and a 2025 meta-analysis by Ghibellini and Meier found no general memory advantage for unfinished tasks, while the urge to resume them (the Ovsiankina effect) held up better. The safe design reading: people do feel a pull to complete open, visible tasks, especially ones they chose and that are nearly done (see goal-gradient in chapter 17). Do not rely on a memory effect to make people return.

In the wild

• Instagram tab barMeta, November 2020
  Reels took the center tab, Shop took the activity tab, and the new-post button moved to the top. Many users reported tapping the wrong tab out of habit. A redesign that prioritizes business goals over habit pays in frustration.
• Safari 15 for Mac (beta)Apple, 2021
  The beta merged tabs and the address bar into a compact design; after feedback, Apple made the traditional separate layout available again before release. Even a well-reasoned change can collide with years of habit; keep an escape hatch.
• Microsoft Office "personalized menus"Office 2000
  Menus hid rarely used items and reordered by use. The adaptive menus were widely disliked and were dropped with the 2007 Ribbon. Adaptive layouts break spatial memory; stability beats cleverness.

The time to choose among equally likely options grows with the logarithm of how many there are. It is a real law, but only for a narrow kind of choice; for scanning lists and making real-life decisions, other things matter more.

In 1952 W. E. Hick and in 1953 Ray Hyman showed that when people must respond to one of several lights by pressing the matching key, reaction time rises by a constant amount each time the number of equally likely alternatives doubles. The usual form is T = a + b · log₂(n + 1). Going from 2 to 4 options adds about as much time as going from 4 to 8, or 8 to 16. The log means the cost grows slowly: doubling the options does not double the time.

When Hick's law applies, and when it does not

Hick's experiments measured a specific situation: options are equally likely, the person knows them all, and the mapping from stimulus to response is learned. Interfaces often differ:

• Unfamiliar lists are searched, not decided. When you do not know where an item is, you scan, and scan time grows roughly linearly with the items you must look at. Sorting, grouping and search cut this.
• Familiar, ordered lists become log-like again. Landauer and Nachbar (1985) found menu selection in well-ordered menus (alphabetical or numeric) fit a log relationship: you can jump to roughly the right place.
• Practice flattens it. Experts with a fixed layout pick by location from memory.
• Unequal likelihoods change everything. If one option is chosen 80% of the time, making it the obvious default removes the decision for most people.

So the practical reading is not "fewer options are always better". It is: reduce the number of options that must be considered at each moment, by defaults, grouping, progressive disclosure and good ordering, while keeping everything findable.

In the wild

• Google homepageGoogle, 1998–
  One field, two buttons, and most people use only the field and Enter. The famous minimalism is a near-zero decision cost at the moment of arrival.
• iOS share sheetApple
  Suggested people and recent apps come first; the long list of actions is below and editable. Likelihood ordering beats fewer options.
• Apple Pay / Google Pay buttonsApple, Google
  At checkout, one prominent button replaces choosing card, address and shipping separately. A smart default collapses many choices into one.

The time to hit a target depends on how far away it is and how big it is. Make frequent targets large and close, use screen edges and corners, and put touch targets where thumbs can reach.

Paul Fitts, a US Air Force psychologist, published the law in 1954 from experiments where people tapped back and forth between two metal plates. Movement time rises with an "index of difficulty" based on the ratio of distance to width. The formulation most used in HCI today is MacKenzie's (1992): MT = a + b · log₂(D/W + 1), where D is distance to the target's center and W is the target's width along the direction of movement. It is one of the most robust laws in human-computer interaction; it holds for mice, trackpads, fingers, styluses and even eye-gaze pointing, with different constants.

Edges and corners: infinitely large targets

On a desktop, the pointer stops at the screen edge. A target that touches the edge is effectively infinitely deep in the direction of movement: you can throw the mouse at it without aiming. Corners are best of all, capped in two directions. Bruce Tognazzini, Apple's early human interface evangelist, popularized this, calling the four corners the easiest targets on screen after the pixel under the cursor. It is why the Mac menu bar sits at the very top edge of the screen, and why hot corners work. A window-attached menu bar a few pixels below the edge loses the effect.

Thumbs: Fitts on a phone

On touch screens, distance is measured from where the thumb rests, and target width is limited by the finger's contact area. Steven Hoober's 2013 observation study of 1,333 people using phones in public found about 49% using one hand, 36% cradling the phone, and 15% two-handed. Josh Clark's "thumb zone" diagrams (in Designing for Touch, 2015) mapped what one-handed users can reach comfortably. Today's phones are tall, so the top of the screen is the hardest place to reach, which is why many apps have moved key actions and even search fields toward the bottom.

Minimum sizes

The visible shape can be smaller than the hit area. A 20 px icon can sit in a 44 px invisible tap target. Spacing matters too: two 44 px targets touching each other still cause mis-taps; leave at least 8 px between destructive and safe actions.

In the wild

• macOS menu barApple, 1984–
  Always at the top screen edge, regardless of window position. Infinite-depth targets for the most frequent commands.
• Safari on iPhoneApple, iOS 15, 2021
  Moved the address bar to the bottom by default (with an option to restore the top). Thumb reach on tall phones; habit and Fitts in tension, resolved with a setting.
• Pie and radial menusDon Hopkins and others, 1980s–; games
  Every item is equally close to the pointer and large in its angular direction. The geometry of Fitts's law, used directly.

Below 0.1 s a response feels instant. Up to 1 s the user's flow of thought survives. Around 10 s, attention leaves. Your job is to keep real latency low and, where you cannot, to make waiting feel short and understood.

In 1968, Robert B. Miller at IBM published "Response time in man-computer conversational transactions", estimating how long people tolerate different kinds of waits. Jakob Nielsen distilled it in Usability Engineering (1993) into three limits that have survived every hardware generation because they are about people, not machines:

• 0.1 second: the limit for feeling that the system reacts instantly, as if you are manipulating the object directly. Typing, dragging, toggles, hover states must be here.
• 1 second: the limit for the user's flow of thought to stay uninterrupted. They notice the delay but do not lose the thread. No special feedback strictly needed, though a light busy indicator after ~300–500 ms helps.
• 10 seconds: the limit for keeping attention on the dialogue. Beyond it, people start doing other things, so show percent-done progress, let them work elsewhere, and tell them when it finishes.

The Doherty threshold

In 1982 Walter Doherty and Ahrvind Thadani at IBM published "The Economic Value of Rapid Response Time", reporting that when computer response time dropped below about 400 ms, user productivity rose sharply, as people and computer settled into a fast conversational rhythm. Be honest about the source: it is an industry paper from the mainframe era, measuring programmers' transactions, not a modern controlled experiment. The precise 400 ms number should not be treated as a law. The direction it points has been confirmed many times since: faster responses keep people engaged and productive, and every extra few hundred milliseconds on a frequent action adds up.

Perceived speed

When you cannot make it faster, you can still make it feel faster and more trustworthy:

1. Acknowledge within 0.1 s. The button visibly presses, the row highlights. This is the most important rule; it tells people their action registered.
2. Delay the spinner. Show busy indicators only if the wait passes about 300–500 ms, so fast actions do not flash.
3. Optimistic updates. Show the result immediately and sync in the background, with a clear and recoverable failure state.
4. Skeletons and progressive loading for content screens, so structure appears before data.
5. Determinate progress for long tasks: a bar that moves steadily feels shorter than a spinner. Tell people what is happening ("Importing 1,240 notes").
6. Let people leave. Above 10 s, run it in the background and notify on completion.

The craft of loading motion, skeletons and timing curves is in Motion & Feel.

Parkinson's law and time limits

C. Northcote Parkinson's 1955 essay in The Economist opened: "Work expands so as to fill the time available for its completion." It was satire about bureaucracy, not science. As a design heuristic it suggests that people take as long as a task seems to allow: an open-ended form invites over-thinking, while a clear, small scope ("Add a title, you can change it later") gets people moving. Use it gently: real time limits (countdowns, expiring offers) add stress and are a common dark pattern.

In the wild

• LinearLinear, 2019–
  Built around a local-first sync engine so most actions update the UI instantly and sync afterwards. Speed is a core part of its identity. Latency is a design decision made in the architecture.
• Instagram uploadsInstagram
  Has been widely reported to start uploading photos in the background while you are still writing the caption, so "Share" feels instant. Do work before it is asked for.
• macOS Finder copyApple
  Long copies show a determinate progress bar with time remaining in a separate, non-blocking window. Above 10 s, let people get on with other things.

People mostly run on a fast, automatic mode of thinking and switch to slow, deliberate thought only when they must. They pick the first option that seems good enough, and they would rather start doing than read instructions. Design for that person, not for an ideal careful one.

System 1 and System 2

In Thinking, Fast and Slow (2011), Daniel Kahneman popularized a distinction from psychologists Keith Stanovich and Richard West: System 1 is fast, automatic, effortless and associative (reading a familiar word, recognizing a face, knowing that 2 + 2 = 4). System 2 is slow, effortful, deliberate and limited (computing 17 × 24, comparing two insurance plans). System 1 is always on and generates impressions; System 2 is lazy and only steps in when something feels wrong or hard.

Kahneman described these as useful fictions, characters in a story about the mind, not brain regions. The book also included some priming studies that later failed to replicate, which Kahneman publicly acknowledged. The core two-mode idea, and the heuristics and biases research with Amos Tversky, are on much firmer ground.

Satisficing: good enough wins

Herbert Simon, who won the 1978 Nobel Prize in economics, coined "satisficing" (satisfy + suffice) in 1956: real decision-makers with limited time and information do not optimize; they pick the first option that meets their threshold. Steve Krug turned this into one of the best-known observations in web usability in Don't Make Me Think (2000): users do not read pages and choose the best link; they scan, pick the first reasonable option, and if it is wrong, press Back. Guessing is cheap and usually works, so it is rational.

Design consequence: the first plausible thing people see will be clicked. Make it the right thing, and make recovery (Back, Undo) cheap when it is not.

The paradox of the active user

John Carroll and Mary Beth Rosson (1987) studied people learning office software and found that they skipped manuals and tutorials to start doing real work immediately, even when reading first would have saved time. They called it a paradox: people would be more productive if they invested in learning, but their motivation to get something done right now wins. They also tend to stick with the methods they first learned, even inefficient ones ("production bias" and "assimilation bias").

In the wild

• GitHub "delete repository"GitHub
  Asks you to type the repository's name to confirm. Deliberate friction that forces System 2 for one rare, destructive action.
• Apple Notes, iA WriterApple; Information Architects
  Open straight into a blank document ready for typing. The active user gets to work instantly; there is nothing to learn first.
• Figma's interactive file onboardingFigma
  Has used playground files where you learn tools by doing small tasks rather than reading. Learning by doing matches the active user.

How you present a choice changes what people choose: the default, the first number they see, the wording, and whether outcomes look like gains or losses. You cannot design a neutral choice, so design an honest one.

Richard Thaler and Cass Sunstein called this "choice architecture" in Nudge (2008): whoever designs the form shapes the decision. As the designer of your product, you are a choice architect whether you like it or not. This chapter covers the effects with the best evidence, plus one famous one with shaky evidence.

Defaults and status quo bias

William Samuelson and Richard Zeckhauser (1988) showed that people disproportionately stick with the current state of affairs. Eric Johnson and Daniel Goldstein's 2003 paper "Do Defaults Save Lives?" compared organ-donor consent across European countries: in countries where people were donors unless they opted out, effective consent rates were close to 100%; in countries requiring people to opt in, they were far lower (Germany around 12%, against Austria's near-universal rate). Same people, same values, different default.

Defaults win for three reasons: they take no effort (System 1), they look like a recommendation ("the designer chose this for a reason"), and changing them feels like taking a risk you could regret. For software, this means your defaults are the product for most people. Most users never open settings.

Anchoring

In Tversky and Kahneman's 1974 paper "Judgment under Uncertainty: Heuristics and Biases", people spun a wheel of fortune rigged to land on 10 or 65, then estimated the percentage of African countries in the UN. Those who saw 65 gave much higher estimates. An arbitrary first number pulls later judgments toward it. In software: the first price people see sets their sense of what is expensive, the first suggested value in a field (a tip, a donation amount, a goal) pulls choices toward it, and the "was ₹ 2,999" next to a sale price works on anchoring. Honest uses: suggest sensible starting values. Dishonest uses: invented "original" prices.

Framing

Tversky and Kahneman's 1981 "Asian disease" problem described the same programs as "200 people will be saved" or "400 people will die" (out of 600). People preferred the sure option in the gain frame and the gamble in the loss frame. Same facts, different choice. In interfaces, "95% uptime" and "down 18 days a year" describe the same thing; "Free for 14 days" and "₹ 499 charged on 15 March" describe the same trial. Honest framing gives people the frame that helps them decide well, usually both numbers.

Loss aversion

Prospect theory (Kahneman and Tversky, 1979) found that losses feel roughly twice as heavy as equivalent gains. Some researchers (Gal and Rucker, 2018) argue the effect is smaller and more context-dependent than usually claimed, but loss-framed messages do reliably catch attention. This is why "You'll lose your 120-day streak" is so powerful, and why it is so easy to abuse. Use loss framing to warn about real losses (unsaved work, data deletion), not to manufacture anxiety.

Choice overload: the jam study and what came after

Sheena Iyengar and Mark Lepper's 2000 paper described a tasting booth in a California supermarket with either 24 or 6 jams. More people stopped at the large display, but of those who stopped, about 30% bought a jar when there were 6 options, against about 3% with 24. It became one of the most cited findings in design.

Then the replications came. Benjamin Scheibehenne, Rainer Greifeneder and Peter Todd's 2010 meta-analysis of 50 experiments found an average choice-overload effect of virtually zero, with large variation between studies. Alexander Chernev and colleagues' 2015 meta-analysis found the effect does appear, but reliably only under certain conditions: when the choice set is complex, the decision task is difficult, people are unsure of their preferences, or they want to minimize effort. So the honest lesson is not "fewer options sell more". It is: make options easy to compare, help people who do not know what they want, and do not make a hard decision harder with clutter.

In the wild

• App Tracking TransparencyApple, iOS 14.5, 2021
  Turned cross-app tracking from a buried default into an explicit, plain-worded choice per app. Changing a default and its framing changed behavior across an entire industry.
• Retirement auto-enrollmentUK workplace pensions, from 2012
  Employees are enrolled by default and can opt out; participation rose sharply. The textbook case of a default in the user's own long-term interest.
• Basecamp pricing37signals
  Has long used very few, simply described plans with flat pricing. Easy comparison removes the conditions under which choice overload appears.

People remember an experience by its most intense moment and its end, not its average. They speed up as a goal gets close, push harder when they feel they have a head start, notice the one thing that differs, and judge good-looking things as easier to use.

The peak-end rule

In a 1993 study by Kahneman, Fredrickson, Schreiber and Redelmeier, people held a hand in painfully cold water (14°C) for 60 seconds, and in another trial for 60 seconds followed by 30 more seconds at a slightly warmer, still unpleasant 15°C. Asked which to repeat, most chose the longer trial: more total pain, but a better ending. A 1996 study by Redelmeier and Kahneman of colonoscopy patients found that memories of the procedure tracked the worst moment and the final moments, not its length ("duration neglect").

For products: the remembered experience of your app is built from peaks (the best and the worst moments) and endings (how a task, session or subscription ends). A smooth hour can be remembered as awful because of one lost draft. A frustrating setup can be redeemed by a delightful first result.

Goal-gradient and endowed progress

Clark Hull observed in 1932 that rats in a maze ran faster as they got closer to food. Ran Kivetz, Oleg Urminsky and Yuhuang Zheng (2006) found the same in people: customers with a coffee-shop loyalty card bought coffee more often as they got closer to the free drink. Joseph Nunes and Xavier Drèze (2006) then gave car-wash customers loyalty cards needing 8 stamps, or cards needing 10 stamps with 2 already stamped. The task was identical (8 more washes), but 34% of the "head start" group completed it against 19% of the others. This is the endowed progress effect: progress that feels already begun motivates.

The isolation (von Restorff) effect

Hedwig von Restorff reported in 1933 that an item that differs from the others in a list is remembered better. It is the memory side of preattentive pop-out (chapter 4). Designers use it to highlight the recommended plan, the primary button, the one new item. It only works when one thing differs, and the effect is stronger for things that are also meaningful.

The aesthetic-usability effect

Masaaki Kurosu and Kaori Kashimura (1995) asked people at Hitachi to rate 26 ATM interface layouts for beauty and for expected ease of use. The two ratings correlated strongly. Noam Tractinsky replicated it in Israel in 1997, finding the effect even stronger. Later work showed that attractive interfaces are judged more usable and that people tolerate small problems in them more. Two honest notes. First, perceived usability is not actual usability: beauty can hide problems from your testers. Second, the effect is about first impressions and goodwill; it does not rescue a product that fails at its job. Visual craft is covered in Visual Design.

In the wild

• MailchimpIntuit Mailchimp
  After you send a campaign, the mascot's hand gives you a high five. Sending an email campaign is stressful; the ending is a small celebration. A designed positive end to an anxious task.
• AsanaAsana
  Occasionally shows a celebratory creature flying across the screen when you complete a task (it can be turned off). A playful peak, made optional for those who find it distracting.
• LinkedIn profile strengthLinkedIn
  A meter shows how complete your profile is and suggests the next step. Goal-gradient and endowed progress in one widget; useful when the goal is the user's, nagging when it is not.

Every task has complexity that cannot be removed, only moved, so move it into the software. Accept messy input and produce clean output. Follow the conventions people learned elsewhere. Prefer the simplest design that works, and focus on the few features that do most of the work.

These five are rules of thumb, not experimental findings. They earn their place because each names a trade-off that designers face weekly.

Tesler's law: conservation of complexity

Larry Tesler (who worked at Xerox PARC and Apple, and is credited with cut, copy and paste as we know them) argued in the mid-1980s that every application has an inherent amount of complexity that cannot be removed. The only question is who deals with it: the user or the developer. He argued that an engineer should spend an extra week to save each of millions of users a minute a day.

Postel's law: liberal in, conservative out

Jon Postel wrote the robustness principle into early internet protocol specifications (RFC 760 in 1980 and RFC 793, TCP, in 1981): "be conservative in what you do, be liberal in what you accept from others." For interfaces: accept phone numbers with or without spaces, dates in several formats, pasted text with stray whitespace, emails with capital letters; then store and display them in one clean format. A caution from the protocol world: being too liberal can hide errors and create security issues, so be liberal about format, strict about meaning (do not guess when the input is genuinely ambiguous).

Jakob's law: people spend most of their time elsewhere

Jakob Nielsen observed that users spend most of their time on other sites and apps, so they expect yours to work the same way. In perception terms (chapter 1), conventions are predictions the brain already has. Breaking them has a cost every time, and must buy something worth more. Conventions to keep unless you have a strong reason: logo links home, search is a magnifying glass, the shopping cart is top-right on the web, swipe-back on iOS, Cmd/Ctrl+Z undoes, Esc closes, a gear means settings. Platform conventions are covered in Systems & Platforms.

Occam's razor and Pareto

Occam's razor, from the 14th-century philosopher William of Ockham, says not to multiply entities beyond necessity: among designs that solve the problem equally, prefer the one with fewest parts. Each extra element, setting or mode adds a little perceptual, memory and decision load from every chapter of this guide. The Pareto principle, named by quality expert Joseph Juran after the economist Vilfredo Pareto's observation of income distribution, says that roughly 80% of effects come from 20% of causes. In products it is often true that a small set of features accounts for most use, and a few bugs cause most crashes. Treat the 80/20 as a shape, not a precise ratio.

In the wild

• FantasticalFlexibits, 2011–
  Known for natural-language event entry that shows the parsed result live as you type. Tesler's law: the parser carries the complexity, and the live preview keeps the user in control.
• Stripe API and CheckoutStripe
  Absorbs card networks, 3-D Secure, local payment methods and tax rules behind a small surface. A business built on moving complexity from users into software.
• Google SheetsGoogle
  Follows Excel's formulas, shortcuts and layout closely. Jakob's law as strategy: switching costs fall when nothing needs relearning.

Lasting motivation comes from feeling in control, getting better, and being connected to others. A behavior happens when motivation, ability and a prompt meet at the same moment. Make the right action easy and well-timed; do not try to pump motivation with tricks.

Self-determination theory

Edward Deci and Richard Ryan developed self-determination theory from the 1970s onward (major statements in 1985 and 2000). It is one of the most researched theories of human motivation. It holds that people have three basic psychological needs, and that activities which satisfy them are pursued for their own sake (intrinsic motivation), while activities that thwart them feel like chores. A famous early finding by Deci (1971) was that paying people for an activity they already enjoyed could reduce their interest in it afterwards: external rewards can crowd out intrinsic motivation when they feel controlling.

Fogg Behavior Model: B = MAP

BJ Fogg at Stanford proposed in 2009 that a behavior (B) happens when three elements converge at the same moment: Motivation (M), Ability (A), and a Prompt (P; he originally called it a "trigger"). If any is missing, the behavior does not happen. Motivation and ability trade off: a very easy behavior needs little motivation, a hard one needs a lot. The prompt only works when the person is above the "action line". His 2019 book Tiny Habits applies it to personal habit building: make the behavior tiny, anchor it to an existing routine, celebrate.

The ethics of behavior design

The same model that helps someone build a reading habit can be used to keep people scrolling at 2 a.m. Several products built on persuasive design have been criticized for exactly that, and some of their designers have since spoken publicly against attention-maximizing design. A practical test for your product:

• Whose goal is it? Prompts that help people do what they set out to do are help. Prompts that create new wants on your schedule are manipulation.
• Would they thank you later? Imagine the user looking back at a month of your prompts. Would they say "that helped" or "that hooked me"?
• Can they turn it off easily? If not, you do not believe they would choose it.

This is developed fully in Honest & Calm.

In the wild

• DuolingoDuolingo, 2012–
  Tiny lessons (ability), daily reminders (prompt), visible progress (competence), and streaks. It added "streak freezes" so a missed day need not erase progress. A strong behavior-design product, and a live debate on where streak pressure stops being helpful.
• StravaStrava, 2009–
  Kudos and clubs connect people around exercise they already chose to do. Relatedness that reinforces an intrinsic goal.
• ObsidianObsidian, 2020–
  Notes are plain Markdown files on your own disk, and almost everything is configurable. Deep autonomy as a core value attracts deeply loyal users.

People respond to a product on three levels at once: an instant gut reaction to how it looks, a feeling from how it works in use, and a considered judgment of what it means to them. Delight is real but fragile; trust is slow to build and quick to lose.

Norman's three levels

In Emotional Design (2004), Don Norman, who had been known for championing usability over beauty, argued that attractive things work better, and offered a model of three levels of processing:

• Visceral: the instant, pre-conscious reaction to appearance, sound and feel. It is where the aesthetic-usability effect lives (chapter 17), and where your app icon, first screen and typography do their work. Studies by Gitte Lindgaard and colleagues (2006) found people form reliable impressions of web pages' visual appeal in as little as 50 milliseconds.
• Behavioral: the pleasure and frustration of use: control, speed, predictability, the feel of scrolling and typing. This is most of this guide.
• Reflective: conscious thought about the product: what owning or using it says about you, the story you tell others, memories of using it. It is shaped by brand, values and peak-end memories.

Delight versus satisfaction: the Kano pointer

Noriaki Kano's model (1984) sorts features by how they affect satisfaction. Basic features cause dissatisfaction when missing but no delight when present (sync that works, undo). Performance features increase satisfaction roughly in proportion (faster search). Delighters are unexpected and cause outsized pleasure, but people do not miss them when absent. Over time, delighters decay into expectations: what delighted in 2010 is basic in 2026. The practical rule: delight sits on top of a working foundation, never instead of it. Aarron Walter's Designing for Emotion (2011) expressed this as a hierarchy: functional, reliable, usable, then pleasurable. Kano is applied to prioritizing features in Problems Worth Solving.

Trust signals

Trust is a judgment about whether a product will act in your interest when you are not watching. The Stanford Web Credibility Project, led by BJ Fogg (2002), asked over 2,500 people to compare websites' credibility; the most frequently mentioned factor was the "design look" of the site. That is System 1 judging competence from polish. But the visual impression is only a first filter. Lasting trust comes from behavior over time:

In the wild

• Things 3Cultured Code, 2017
  Won an Apple Design Award; known for refined visuals (visceral), a smooth, predictable feel in use (behavioral) and a reputation for craft that users are proud to recommend (reflective). All three levels pulling the same way.
• Stripe status and docsStripe
  Public status page, precise error codes, and detailed changelogs for API versions. Transparency and reliability as trust signals to developers.
• Braun and Dieter RamsBraun, 1955–1995
  Products whose calm visual order (visceral) matched their honest function (behavioral) and became objects of pride (reflective). The physical template for calm software; see Breakdowns: Objects.

A law is useful only if you know what it predicts, how strong the evidence is, and what to do differently on Monday. This table gives all three.

Evidence ratings: Strong means decades of replicated lab or field results; Moderate means real effects whose size or conditions vary; Heuristic means a wise rule of thumb, not an experimental finding; Contested means the famous version has failed to replicate or is much narrower than claimed.

How to use this table without becoming a law-quoter

1. Start from an observed problem, not a law. "People miss the error" comes first; "change blindness" is the explanation that suggests fixes.
2. Check the evidence column before betting a design on a law. Heuristics are prompts for thinking, not proofs.
3. Expect laws to conflict. Fitts wants big buttons; Occam wants fewer parts; Jakob wants convention; delight wants surprise. Your judgment weighs them for this user and task.
4. Verify with people. A five-person test beats a citation. See Testing & Iteration.

In the wild

• Laws of UXJon Yablonski, website 2017, book 2020
  Popularized a curated set of psychology principles for designers with clear summaries. A great index; read the original studies behind the laws you rely on most.
• Nielsen's 10 heuristicsJakob Nielsen, 1994
  Several heuristics (visibility of status, recognition over recall, error prevention) are the laws in this guide turned into review questions. Laws become useful as questions you ask of a screen.
• Apple Human Interface GuidelinesApple
  Target sizes, feedback, consistency and platform conventions encode Fitts, response time and Jakob's law into concrete rules. Platform guidelines are applied psychology; learn the "why" so you know when they bend.

Wrap-up

You now have a working model of the person on the other side of the screen: a visual system that groups and predicts, an attention that holds one thing, a memory that holds about four, a hand that obeys Fitts, a decision-maker who satisfices and follows defaults, and a human who wants autonomy, competence and connection and remembers peaks and endings. Every later guide builds on this. Typography is perception applied to reading; motion is common fate and response time; words are recognition and framing; ethics is what you do with all of it.

Practice assignment for the next two weeks

About 30 minutes a day. Each day produces one small artefact in a single "Mind notes" document.

Reading list

• The Design of Everyday Things (revised edition, 2013)
  Don Norman · mental models, signifiers, the gulfs of execution and evaluation; the foundation for this whole guide.
• Designing with the Mind in Mind (3rd edition, 2020)
  Jeff Johnson · the best single book on perception, attention and memory written for interface designers.
• The Humane Interface (2000)
  Jef Raskin · locus of attention, habituation and modes, argued with rigor and strong opinions.
• Thinking, Fast and Slow (2011)
  Daniel Kahneman · System 1 and 2, heuristics and biases, peak-end. Read the priming chapter with the replication caveat in mind.
• Don't Make Me Think, Revisited (2014)
  Steve Krug · satisficing and scanning in plain language; a two-evening read.
• Laws of UX (2020)
  Jon Yablonski · compact chapters on the most-cited laws; pair it with this guide's evidence ratings.
• Information Visualization: Perception for Design (4th edition, 2020)
  Colin Ware · the deep reference on preattentive features and visual search.
• Emotional Design (2004)
  Don Norman · the visceral, behavioral and reflective levels.
• Flow (1990)
  Mihaly Csikszentmihalyi · the conditions for absorbed, optimal experience.
• Tiny Habits (2019)
  BJ Fogg · B = MAP in practice; read it alongside the ethics questions in chapter 19.
• "The magical number 4 in short-term memory" (2001)
  Nelson Cowan, Behavioral and Brain Sciences · the modern capacity estimate and why 7 is not it.
• "The Magical Number Seven, Plus or Minus Two" (1956)
  George A. Miller, Psychological Review · short, witty, and very different from how it is quoted.
• "F-Shaped Pattern of Reading on the Web: Misunderstood, But Still Relevant" (2017)
  Kara Pernice, Nielsen Norman Group · the corrected story of the F-pattern and the other scanning patterns.
• Nudge (2008; final edition 2021)
  Richard Thaler and Cass Sunstein · choice architecture and defaults.
• Calm Technology (2015)
  Amber Case · principles for technology that respects attention.

Next: put the user's mind to work on the right problem in Problems Worth Solving, or go straight to the perception-heavy guides, Visual Design and Typography.