UX / UI designer interview questions

A PM asks for a “modern settings page.” First move?

The strongest answer: Name the job a person is trying to finish and what is in the way — I do not open a moodboard first

Why it works: Exactly. The job first. Trends, copies, and invented scores skip the person.

• Frame the job before pixels.
• Do not start in a trend file.
• Do not copy a competitor as the brief.
• No invented usability score.

What must a research session have before you book people?

The strongest answer: A question you could be wrong about — not “validate our mock” as the only goal

Why it works: Exactly. A falsifiable question. Compliment scripts and promised 90% are theatre.

• Start with a question you could lose.
• Do not only validate a mock.
• Teammates ≠ users.
• No promised preference %.

When are lo-fi wireframes the stronger artefact?

The strongest answer: When we are still arguing flow, states, and the job — polish is how you freeze the wrong path

Why it works: Exactly. Fidelity matches the argument. Motion reels freeze the wrong path with extra charm.

• Lo-fi while flow and states are open.
• Polish freezes a path.
• Polish ≠ senior.
• Do not start in a motion reel.

How do you set type on a page so the hierarchy reads without shouting?

The strongest answer: A small scale — a few sizes with clear steps between them — plus weight, colour and space to separate the levels, so two levels never look nearly the same

Why it works: Right — few steps, big enough to be deliberate, with weight and space doing work alongside size.

• A few sizes, clear steps.
• Define the scale first.
• Use weight, colour and space too.
• Never fix layout by shrinking text.

What is Figma for, and what is it not?

The strongest answer: A place to try and share decisions — it is not the research, the a11y, or the shipped UI by itself

Why it works: Exactly. A tool. Done-in-Figma, missing states, and version folklore are how you fail the handoff.

• Figma shares decisions.
• It is not the research or the ship.
• Do not leave states to engineering.
• Do not hide unofficial versions.

The list is empty on day one. What do you design?

The strongest answer: A next step that starts the job — not a cute illustration and a shrug, and not fake rows for the screenshot

Why it works: Exactly. A next step. Fake rows and contact-grabs are how you teach distrust.

• Empty = a next step.
• Not only an illustration.
• Do not fake rows for the shot.
• Do not make a contact-grab the only button.

When do you think about accessibility?

The strongest answer: In the first wire — labels, order, contrast, and a real control — I do not “add a11y” in week eight

Why it works: Exactly. First wire. Week-eight, legal-only, and aria-on-div are how you retrofit a mess.

• A11y from the first wire.
• Labels, order, contrast, native controls.
• Not a week-eight phase.
• Aria-on-div is not done.

What is a mental model, and what goes wrong when the interface disagrees with one?

The strongest answer: It is what a person already expects this thing to do; when the UI disagrees they hesitate, undo, or blame themselves — so I either match the expectation or teach it plainly

Why it works: Right — expectation first, and where you must break it, say so in the interface rather than in a help centre.

• Match what people already expect.
• Ask what they expect first.
• Match it, or teach it visibly.
• Never patch it with a tooltip.

How do you start a critique?

The strongest answer: With the job and the question you want help on — not “be nice” and not a roast of the typeface

Why it works: Exactly. Job and question. Font roasts, silence, and votes skip the work.

• State the job and the question.
• Critique the path, not the person.
• Do not roast the type first.
• Do not vote on who is best.

What must a handoff include besides the happy frame?

The strongest answer: Empty, error, loading, permissions, and the copy — I do not send a pretty frame and vanish

Why it works: Exactly. States and words. Links, gifs, and pop are how week two surprises arrive.

• Handoff: empty, error, loading, permissions, copy.
• Do not vanish after a pretty frame.
• A link ≠ a handoff.
• “Pop” is not a spec.

The main navigation reads Hub, Studio and Insights. What do you check before you defend it?

The strongest answer: Whether people can predict what sits behind each one — I test the labels against the words users already say, and rename before we brand the navigation

Why it works: Right — a nav label’s only job is prediction, and the words people already use win.

• Labels must predict the content.
• Use the words users say.
• Test with people, not the team.
• Rename before you brand it.

Growth wants a pre-ticked marketing box buried in the signup. What do you do?

The strongest answer: No — that is a dark pattern; I will design an honest opt-in — I do not “just ship the tick”

Why it works: Exactly. Refuse the pattern. Burials and win-if-worse A/Bs are how you become the case study.

• Refuse pre-ticked consent.
• Design an honest opt-in.
• Do not bury the untick.
• Conversion ≠ permission to trick.

You have one week. How do you plan research?

The strongest answer: One question, a few of the right people, a synthesis you will use Monday — I do not book twelve methods to look thorough

Why it works: Exactly. One question, usable Monday. Buffets, leading surveys, and Slack-as-users skip the week.

• One question, right people, Monday artefact.
• Do not buffet methods.
• Do not lead the survey.
• Slack ≠ users.

Your spacing values are 6, 7, 10, 13 and 18 px. What do you do, and why does it matter?

The strongest answer: Snap them to a small scale built on one step, such as 4 or 8 — arbitrary values cannot be reproduced, reviewed or tokenised, and the page reads as noise instead of rhythm

Why it works: Right — a scale is what turns spacing into a decision you can name, hand over and reuse.

• Spacing is a scale, not a feeling.
• Pick one step, build the scale.
• Snap every gap onto it.
• Tokenise the scale.

What belongs in a problem frame besides a user quote?

The strongest answer: The job, the constraint, and what we will not solve — a quote alone is a poster

Why it works: Exactly. Job, constraint, non-goal. Posters, unread maps, and invented 9.2s are décor.

• Job + constraint + non-goal.
• A quote is not a frame.
• Do not write an unread novel map.
• No invented pain score.

In review someone says your wireframe “feels random”, even though every gap came from the spacing scale.

What is probably wrong?

The strongest answer: Equal gaps everywhere, so nothing reads as a group — I tighten the space inside a group and open it between groups, and let proximity work before borders do

Why it works: Exactly — relationships come from relative space first; boxes are the fallback, not the tool.

• Space carries grouping.
• Tighten inside, open between.
• Try proximity before borders.
• Squint at it to check.

When do you raise fidelity?

The strongest answer: When the argument has moved to hierarchy, type, or a system token — I do not polish to end a flow fight

Why it works: Exactly. Match the argument. Bored-polish and polish-to-trap are politics.

• Raise fidelity when the argument needs it.
• Do not polish to end a flow fight.
• Do not polish because someone is bored.
• Do not trap stakeholders with polish.

You are specifying one button. Which states do you hand over?

The strongest answer: Default, hover, focus, pressed, disabled and loading — and what disabled means here, because a control that does nothing with no explanation is a dead end

Why it works: Exactly — and focus is not optional: it is the only state a keyboard user can see.

• Six states for every control.
• Default, hover, focus, pressed.
• Disabled and loading too.
• Say what disabled means here.

You need three button states. How do you build them in Figma?

The strongest answer: As variants of one component — I do not draw three snowflakes that will drift

Why it works: Exactly. Variants. Snowflakes, hidden pages, and eng-invents-hover are how the system dies.

• Use component variants.
• Do not draw snowflakes.
• Do not hide states in a junk page.
• Do not leave hover to engineering.

A settings screen shows forty controls in one flat list, and support says people cannot find the two they came for.

How do you cut the load without removing anything?

The strongest answer: Group by the job, give the rare controls sensible defaults and collapse them, and put the two people come for first — the same forty settings, fewer decisions at once

Why it works: Exactly — load is about how many choices are live at once, not how many exist.

• Fewer live choices, not fewer features.
• Group by the job.
• Default and collapse the rare ones.
• Surface what people came for.

A system button exists. You want a slightly different one. First move?

The strongest answer: Use the system, or propose a change with a reason — I do not quietly fork a one-off

Why it works: Exactly. Use or propose. Quiet forks and campaign sizes are how you get 14 buttons.

• Use the system or propose a change.
• Do not quiet-fork.
• Do not invent a campaign size.
• Taste ≠ a fork licence.

What rule makes critique useful?

The strongest answer: We speak to the work against the job, and the designer names the question — I do not let it become a taste pile-on

Why it works: Exactly. Job and a named question. Loud, only-nice, and emoji votes skip the decision.

• Work against the job.
• Designer names the question.
• No taste pile-on.
• No emoji-vote-and-leave.

A card looks right at the width you drew it, but dragging the frame wider leaves the button stranded in the middle.

What did you forget?

The strongest answer: Constraints — or Auto Layout. A child pinned to the left will not follow the right edge; pin it right, or let a layout with fill handle the resize

Why it works: Exactly — resize behaviour is something you specify in the file, and it is half of what handoff reads.

• Resize behaviour is designed, not assumed.
• Set constraints on every child.
• Prefer Auto Layout for lists and bars.
• Drag the frame before you hand over.

What does a usability test need besides a prototype?

The strongest answer: A task in their words, a think-aloud, and notes on where they stall — I do not pitch the design while they try

Why it works: Exactly. Their task, their stalls. Pitches, NPS-in-the-lab, and pre-written success rates are theatre.

• Task in their words + stalls.
• Do not pitch during the attempt.
• Do not swap in an NPS.
• No pre-written success %.

Notes live in twelve Drives. What do you fund?

The strongest answer: A home tagged by question and decision, with consent — I do not fund an 80-page toolkit nobody opens

Why it works: Exactly. Findable by decision. Toolkits, bans, and memory are how you re-interview the same pain.

• Store by question and decision.
• Keep consent with the notes.
• Do not fund a shelf toolkit.
• Do not ban notes to stay light.

How do you run a framing workshop that is not a sticky-note party?

The strongest answer: A job, a constraint, and a non-goal on one page they mark — I do not end with 200 stickies and no sentence

Why it works: Exactly. One marked sentence. Sticky parties, colour votes, and invented pain scores are décor.

• One page: job, constraint, non-goal.
• They mark it.
• Do not end in a sticky pile.
• No invented pain score.

What is a layout grid actually for, and when is breaking it the right call?

The strongest answer: It makes independent screens line up and gives handoff a shared vocabulary; I break it deliberately for a full-bleed element or content that will not fit, and I say that I did

Why it works: Right — the grid is a contract between screens and with engineering, so an exception is announced, not silent.

• A grid is a shared contract.
• Set columns, gutters and margins.
• Break it deliberately, say why.
• Note the exception in handoff.

Eight squads ship eight dropdowns. What do you standardise?

The strongest answer: The component, the tokens, and the a11y contract — I do not standardise every campaign illustration

Why it works: Exactly. Behaviour and tokens. Illustration-only, novels, and snowflake-autonomy miss the a11y contract.

• Standardise component + tokens + a11y.
• Let campaign art differ.
• Do not write a visual novel.
• Do not bless eight dropdowns as autonomy.

A catalogue has grown to nine top-level sections, several of which hold two items.

How do you decide between fewer, deeper sections and more, shallower ones?

The strongest answer: By how confidently people predict what is inside: shallow wins while the labels are clear, but nine vague siblings are worse than four obvious ones with a level underneath

Why it works: Right — the cost is a wrong guess, not a click, so test the labels before you pick a shape.

• Predictability beats click count.
• Group by what users seek.
• Merge the vague siblings.
• Tree-test before you ship.

The file has 40 pages named final. What do you change?

The strongest answer: One source of truth, named variants, and archive the rest — I do not add a 41st final

Why it works: Exactly. One truth. 41st, weekly files, and museum-40 are how eng picks the wrong grey.

• One source of truth + variants.
• Archive the rest.
• Do not add another final.
• Do not keep 40 as a museum.

You need a new pattern. How do you add it without a fork farm?

The strongest answer: A proposal, a use, and a path into the library — I do not hide a local component “just for now”

Why it works: Exactly. Propose, use, graduate. Hidden just-for-now and Slack screenshots are forks.

• Propose, use, graduate into the library.
• Do not hide a just-for-now fork.
• Do not wait forever.
• Slack ≠ shipped to the system.

You hand over a desktop frame and a phone frame for the same page.

What does engineering still not know?

The strongest answer: Everything between them: what wraps, what stacks, what truncates, the minimum and maximum widths, and what happens when the text is twice as long after translation

Why it works: Exactly — two frames are two samples; the rules between them are the design, and long content is where they break.

• Specify rules, not only frames.
• Say what wraps, stacks and truncates.
• Give minimum and maximum widths.
• Test with long translated text.

Your segment is a nurse on a shared ward tablet, gloved, working between patients. What changes in the screen you draw?

The strongest answer: Bigger targets, almost no typing, a task that survives an interruption, and nothing personal left on a shared screen — the constraints change the layout, not the poster

Why it works: Right — a persona earns its keep only when its constraints show up as decisions you can point at.

• Turn the person into constraints.
• List device, hands, interruption, privacy.
• Change layout, input and saved state.
• Point at each decision it caused.

Critique has become a compliment circle. What do you change?

The strongest answer: The prompt: a question the work could fail — I do not add a second meeting called “real critique”

Why it works: Exactly. A fail-able question. Shame, cancel, and winner boards kill the room.

• Ask a question the work could fail.
• Do not add a second “real” critique.
• Do not shame.
• Do not run a winner board.

What do you standardise in usability across teams?

The strongest answer: Consent, task style, and how findings reach a decision — I do not standardise every prototype tool

Why it works: Exactly. Consent and a path to a decision. One-sitter, novels, and invent-every-time miss the bar.

• Standardise consent, task style, decision-path.
• Let tools differ.
• Do not require one sitter on every test.
• Do not write a shelf protocol.

A prototype tested beautifully; the built feature is getting complaints.

What did the prototype hide?

The strongest answer: Real data and real waiting — long names, empty accounts, permissions, slow networks, failures — the prototype ran a clean path on content that was chosen to fit

Why it works: Exactly — a prototype tests comprehension and flow, never latency, data shape or permissions.

• Prototypes test understanding, not reality.
• Retest with real data.
• Include waits, errors and permissions.
• Spec what the prototype could not show.

What do you standardise across eight product design teams?

The strongest answer: How a problem is framed, critiqued, and handed off — I do not standardise every Figma plugin

Why it works: Exactly. Frame, critique, handoff. Plugins, central approvers, and novels are queues and shelves.

• Standardise frame / critique / handoff.
• Let plugins differ.
• Do not approve every frame.
• Do not write a shelf novel.

Designers are rewarded only for shipping screens. What breaks?

The strongest answer: They will skip the question — I reward a killed idea and a well-asked study, not only a launch file

Why it works: Exactly. Incentives write the craft. Ship-only, Dribbble-quotas, and screen-counts are how you get junk.

• Reward questions and honest kills.
• Not only launch files.
• Do not quota Dribbble.
• Do not run a screen-count board.

A second brand is coming, and the product already ships light and dark.

How do you set the files up so this does not become four libraries?

The strongest answer: Colour variables in collections with modes — one for brand, one for theme — and components that only ever reference the semantic variable, so switching a mode reskins a screen with no duplicate frames

Why it works: Right — the discipline is that no component holds a raw value; the modes then carry the whole difference.

• One library; modes carry the difference.
• Components reference semantic variables only.
• Collections with brand and theme modes.
• Never hardcode a raw value.

What does a staff designer add to strategy that a slide deck does not?

The strongest answer: Who we will disappoint, in a moment you can see — I do not decorate a strategy we have not chosen

Why it works: Exactly. A visible disappointment. Pretty decks, motion titles, and invented sizes skip the choice.

• Show whom we disappoint, in a moment.
• Do not decorate an unchosen strategy.
• Pretty ≠ framed.
• No invented opportunity size.

What does design-system governance actually decide?

The strongest answer: What gets in, what is deprecated, and the a11y bar — I do not meet to admire a colour

Why it works: Exactly. In, out, bar. Colour-hours, icon stamps, and invented contribution % are theatre.

• In / deprecate / a11y bar.
• Not a weekly colour salon.
• Not a stamp on every icon.
• No contribution percentile.

The team has been through four prototype rounds and the debate has stopped moving.

When do you stop prototyping and ship a thin slice instead?

The strongest answer: When the open questions need real data, load or money to answer — then I ship the smallest honest slice behind a flag, with a way to watch it and to roll back

Why it works: Right — name what a prototype physically cannot answer, then take the cheapest real test with a way out.

• Stop when the method runs out.
• List the open questions.
• Ask which need reality.
• Ship a small, reversible slice.

How do you teach critique so it survives a reorg?

The strongest answer: A written ritual: job, question, decision — I do not rely on one charismatic facilitator

Why it works: Exactly. A ritual you can hand on. Cults, roasts, and tool scores die or injure.

• Write job / question / decision.
• Do not rely on one facilitator.
• Do not roast.
• Do not replace critique with a tool score.

What does accessibility look like as a program, not a poster?

The strongest answer: Gates, training, and a backlog with owners — I do not print a 100 badge

Why it works: Exactly. Gates and owners. Videos, overlays, and invented percentiles are décor.

• Gates, training, owned backlog.
• No 100 badge as proof.
• Overlays ≠ the program.
• No invented WCAG percentile.

Four products, four different main navigations, and the same customer uses three of them.

What do you standardise, and what do you leave to each team?

The strongest answer: A shared vocabulary, the top-level model, and where the global things live — account, search, help, switching product; each team keeps the structure inside its own area

Why it works: Right — standardise the model and the words, not every screen; that is what makes four products feel like one.

• Standardise vocabulary, not every menu.
• Agree the top-level model.
• Fix where global things live.
• Leave inner structure to teams.

What do you standardise in handoffs across the company?

The strongest answer: States, tokens, and a living leftover — I do not standardise every cover page

Why it works: Exactly. States and leftovers. Covers, required humans, and novels are queues and shelves.

• Standardise states, tokens, leftovers.
• Let covers differ.
• Do not sit in every handoff.
• Do not write a shelf novel.

What ethics bar do you set for the design org?

The strongest answer: No dark patterns, no fake outcome numbers, no hidden cancel — a growth target is not a waiver

Why it works: Exactly. Patterns, honesty, cancel. Legal-minimum, A/B-as-ethics, and 6pt PDFs are how you become the case study.

• Ban dark patterns and hidden cancel.
• No fake outcome numbers.
• Legal ≠ enough.
• A/B ≠ ethics.

The company’s design site is a shot gallery. What do you change?

The strongest answer: Cases with problems, leftovers, and credit — I do not hire with a moodboard of shots

Why it works: Exactly. Stories with leftovers. More shots, invented 2×, and hidden leftovers hire the wrong people.

• Publish problems, leftovers, credit.
• Do not hire from a shot gallery.
• No invented 2× impact.
• Do not hide leftovers.