irreplaceable

Irreplaceable › Guides › UX review checklist

Checklist · Design crit and UX audit

The UX review checklist: 102 rules

Every rule the design crit and Crit Gate check a screen against, written out. Use it for a design crit, a UX audit or your own last look before handoff. Each rule says where it applies, why it matters and how severe a miss is.

By Sarit Elisha, founder of Irreplaceable · Updated October 10, 2026

Run all 102 on a screen

Upload a screenshot. 3 free crits with a free account.

How severe is a miss?

Every rule carries a severity, so a review starts with what matters most.

  • BlockerPeople can't finish, or the screen harms them: money, data, consent.
  • MajorMany people will hesitate, misread or make a mistake.
  • MinorFriction or polish. Worth fixing, rarely urgent.

24 rules

Core UX

Basics every screen needs, AI or not.

CORE-01

MajorHierarchy

One primary action, clearly the strongest

When two actions look equally important, people hesitate or pick the wrong one.

Applies to: Any screen with a main task.

CORE-02

MinorHierarchy

The thing people came to see is the first thing they see

A key number or status buried under decoration gets missed.

Applies to: Screens that show a result, total or status.

CORE-03

MinorClarity

Labels say what happens, not 'OK' or 'Continue'

A specific verb ('Pay $412', 'Delete 3 files') tells people the consequence before they commit.

Applies to: Buttons and links.

CORE-04

MajorClarity

Money shows currency, what it includes and what changed

An unexplained total, or one that differs from what was asked for, breaks trust at the worst moment.

Applies to: Totals, prices and amounts.

CORE-05

MajorSafety

Irreversible actions are named as such and confirmed or undoable

People need a last look, or a way back, when a mistake is expensive.

Applies to: Delete, send, pay, publish and other hard-to-reverse actions.

CORE-06

MinorStates

Empty states explain what goes here and how to start

A blank area reads as broken and leaves people without a next step.

Applies to: Lists, searches, dashboards and first-run screens.

CORE-07

MajorStates

Errors say what went wrong and how to fix it, next to the problem

'Something went wrong' leaves people stuck; an error far from its field is missed.

Applies to: Forms and anything that can fail.

CORE-08

MinorStates

Waiting shows progress and what is happening

A spinner with no context feels longer and makes people retry or leave.

Applies to: Anything that takes more than a moment.

CORE-09

MajorAccess

Small or low-contrast text that carries meaning

Fine print that holds a key fact (a fee, a non-refundable term) is easy to miss and reads as hiding it.

Applies to: Text and controls.

CORE-10

MinorForms

Every field has a visible label, not only a placeholder

Placeholders disappear on typing, so people lose track of what each field wants.

Applies to: Inputs.

CORE-11

MinorHeuristics

Every action gets visible feedback

Without a response, people don't know if the tap worked, tap again, and create duplicates or give up.

Applies to: Every action people can take on the screen.

CORE-12

MinorHeuristics

Words match the user's world, not the system's

Internal names, codes and database terms make people translate before they can act.

Applies to: Labels, headings and messages.

CORE-13

MajorHeuristics

A way back, out or undo is always visible

People explore when they know they can return. Without Back, Cancel or Undo they hesitate or get stuck.

Applies to: Screens inside a flow, modals and edits.

CORE-14

MinorHeuristics

The same thing looks and works the same everywhere

Two styles for one action, or one style for two actions, make people relearn the screen as they go.

Applies to: Repeated components and patterns on the screen.

CORE-15

MinorHeuristics

Errors are prevented, not just reported

Constraints, good defaults and pickers stop wrong input before it happens, which beats any error message.

Applies to: Inputs, pickers and destructive controls.

CORE-16

MinorHeuristics

Help is where the question comes up

A hint next to a hard field gets read; a help centre link at the bottom doesn't.

Applies to: Complex fields, settings and first use.

CORE-17

MinorHeuristics

Frequent users have a faster path

Shortcuts, bulk actions and remembered choices keep experts fast without getting in a newcomer's way.

Applies to: Screens people use often or in bulk.

CORE-18

MinorHeuristics

Every element earns its place

Each extra element competes with the ones that matter and slows the main task.

Applies to: Dense screens.

CORE-19

MinorNavigation

People can tell where they are

Without a title, active navigation state or breadcrumb, people lose their place and can't predict where Back goes.

Applies to: Screens inside an app or site.

CORE-20

MinorMobile

The main action is within thumb reach

Actions at the top of a tall phone screen need two hands or a stretch, and get missed or delayed.

Applies to: Primary actions on phone screens.

CORE-21

MinorMobile

The keyboard fits the field and doesn't hide it

Email, number and phone fields need their own keyboard, and the field and its error must stay visible above the keyboard.

Applies to: Text fields on phone screens.

CORE-22

MinorForms

Required and optional are clear, and input formats are forgiving

Guessing which fields are required, or being rejected for a space in a card number, causes most form errors.

Applies to: Forms.

CORE-23

MinorData

Data is easy to scan and compare

Numbers aligned on the right with consistent units, clear headers and sensible sorting let people compare at a glance.

Applies to: Tables, lists and numbers.

CORE-24

MinorLayout

Grouping and hierarchy show the structure

Related things close together and a clear order of headings let people see the structure before they read it.

Applies to: Any screen with more than a few elements.

18 rules · the full guide

AI UX

How AI features act, admit uncertainty, ask for consent and recover. From the Irreplaceable AI UX checklist.

AIUX-01

MajorBefore the AI acts

Show the plan before any costly or shared action

People can't undo what they never saw. A plan turns a surprise into a decision.

Applies to: Agents that book, send, pay or change things.

See it done right and wrong

AIUX-02

MajorBefore the AI acts

Ask one clear question when the request is ambiguous

Guessing on ambiguity causes costly mistakes. Asking five questions every time causes friction.

Applies to: Agents that book, send, pay or change things.

See it done right and wrong

AIUX-03

MajorBefore the AI acts

Let people edit the plan before it runs

Steering before acting beats undoing after.

Applies to: Agents that book, send, pay or change things.

See it done right and wrong

AIUX-04

MinorUndo and control

Give an undo window for anything sent or deleted

Sending and deleting can't be taken back. A few seconds of undo covers most mistakes.

Applies to: Anything AI writes, edits or runs for a while.

See it done right and wrong

AIUX-05

MinorUndo and control

Keep the original when AI rewrites or edits

People need to see what changed and get their own words back.

Applies to: Anything AI writes, edits or runs for a while.

See it done right and wrong

AIUX-06

MinorUndo and control

Let people stop long generations

Control matters most while it's still running.

Applies to: Anything AI writes, edits or runs for a while.

See it done right and wrong

AIUX-07

MinorHonesty about uncertainty

Say “I'm not sure” and show sources when unsure

Honest uncertainty builds trust that lasts. Confident mistakes break it.

Applies to: Answers, summaries and recommendations.

See it done right and wrong

AIUX-08

MinorHonesty about uncertainty

Don't invent precision

Numbers like “92% match” sound scientific but give people nothing to act on.

Applies to: Answers, summaries and recommendations.

See it done right and wrong

AIUX-09

MinorHonesty about uncertainty

Show where every claim came from

A claim people can check is a claim people can trust.

Applies to: Answers, summaries and recommendations.

See it done right and wrong

AIUX-10

MajorConsent and privacy

Ask before remembering anything sensitive

Health, money and family details need consent, not silent memory.

Applies to: Memory, permissions and personal data.

See it done right and wrong

AIUX-11

MajorConsent and privacy

Limit permissions to the task, with an end date

Forever access to everything is a risk that never goes away.

Applies to: Memory, permissions and personal data.

See it done right and wrong

AIUX-12

MajorConsent and privacy

Leave consent boxes unchecked

A pre-checked box isn't consent, and in the EU it isn't legal.

Applies to: Memory, permissions and personal data.

See it done right and wrong

AIUX-13

MinorBe clear it's AI

Say it's an AI, upfront

People have a right to know when they're talking to a machine.

Applies to: Assistants, chatbots, generated content.

See it done right and wrong

AIUX-14

MinorBe clear it's AI

Label AI-generated images and text

People decide differently when they know something is generated.

Applies to: Assistants, chatbots, generated content.

See it done right and wrong

AIUX-15

MinorBe clear it's AI

Don't fake feelings to keep people engaged

Emotional pressure from a machine is manipulation, especially for kids.

Applies to: Assistants, chatbots, generated content.

See it done right and wrong

AIUX-16

MajorWhen things go wrong

Say what happened and what to do next

An error message is a moment to help, not to blame.

Applies to: Errors, refusals and high-stakes moments.

See it done right and wrong

AIUX-17

MajorWhen things go wrong

Show a way to a person when stakes are high

Money, health and legal moments need a human exit.

Applies to: Errors, refusals and high-stakes moments.

See it done right and wrong

AIUX-18

MajorWhen things go wrong

Explain a refusal and offer what you can do

A bare “I can't help” is a dead end.

Applies to: Errors, refusals and high-stakes moments.

See it done right and wrong

50 rules · the full guide

Behavioral

Behavioral science and cognitive psychology: how people notice, decide, remember and act, and the dark patterns that exploit it.

BEH-01

MinorChoice

Few enough choices to decide (Hick's law, choice overload)

Every extra option slows the decision, and too many make people give up or pick at random. Group, recommend one, or hide advanced options.

Applies to: Screens that ask people to choose: plans, options, settings, menus.

BEH-02

MajorChoice

Defaults serve the user, not the business

Most people keep the default. A pre-ticked add-on, newsletter or data-sharing box is a choice made for them, and erodes trust when they notice.

Applies to: Pre-selected options, toggles and checkboxes.

BEH-03

MajorChoice

Anchors and decoys are honest

A crossed-out price that was never charged, or a decoy plan that exists only to push another, is anchoring used against people; regulators treat fake reference prices as deceptive.

Applies to: Prices, plans and comparisons.

BEH-04

MinorLoad

Cognitive load fits one step

Working memory holds only a few things at once. Asking for everything on one screen raises errors and drop-off; split it, prefill it, or ask later.

Applies to: Forms, onboarding and dense screens.

BEH-05

MinorLoad

Recognition over recall

Making people remember a code, a choice or a rule from an earlier step causes mistakes. Show it where it's needed.

Applies to: Anything people must remember from a previous screen.

BEH-06

MinorLoad

Progressive disclosure: the essential first

Showing every detail up front buries the next step. Lead with what most people need and reveal the rest on demand.

Applies to: Settings, advanced features and long content.

BEH-07

MinorMotivation

Progress is visible (goal gradient)

People speed up as they see the end and abandon when they can't tell how far it is. Show steps or progress, and start it a little ahead if a step is already done.

Applies to: Multi-step flows, onboarding and profiles.

BEH-08

MinorMotivation

The end of the experience lands well (peak-end rule)

People judge an experience by its peak and its end. A flat or confusing ending colours the whole flow; confirm what happened and what comes next.

Applies to: Confirmation, success and end-of-flow screens.

BEH-09

MinorMotivation

The ask comes at a moment of motivation and ability

A prompt works when people want to and can act now (Fogg model). Asking at the wrong moment trains them to dismiss it.

Applies to: Prompts to act: upgrades, invitations, reminders.

BEH-10

MinorFraming

Framing doesn't mislead (framing effect, loss aversion)

'90% success' and '10% failure' feel different though they're the same. Loss-framed scare copy pushes decisions people later regret.

Applies to: Risks, fees, results and AI output.

BEH-11

MajorFraming

Social proof is real and relevant

'12 people are looking at this' that isn't true, or reviews from another product, is fake social proof and a recognised dark pattern.

Applies to: Reviews, counts of other users, badges.

BEH-12

BlockerDark patterns

No false urgency or scarcity

Countdown timers that reset and fake 'only 2 left' pressure people into choices; regulators in the EU, UK and US act against them.

Applies to: Timers, stock counts and limited offers.

BEH-13

BlockerDark patterns

Leaving is as easy as joining (no roach motel, no confirmshaming)

Hiding cancel, or wording the decline button to shame ('No, I don't like saving money'), keeps people by friction and guilt, and costs trust and often legal exposure.

Applies to: Cancel, unsubscribe, decline and opt-out paths.

BEH-14

BlockerDark patterns

No hidden costs, sneaking or forced continuity

Fees revealed at the last step, items added to the basket, or a free trial that silently becomes paid exploit commitment and inattention.

Applies to: Checkouts, sign-ups and consent.

BEH-15

MinorHabit

Habit loops respect the user's own goal

Streaks and variable rewards build habits; aimed at engagement for its own sake they turn into compulsion and churn once people notice.

Applies to: Notifications, streaks and engagement loops.

BEH-16

MajorTrust

Automation invites checking, not blind trust (automation bias)

People over-trust confident systems and stop checking. Show sources and uncertainty where a wrong answer costs something.

Applies to: AI suggestions, recommendations and automation.

BEH-17

MinorAttention

Frequent actions are big and close (Fitts's law)

The time to hit a target grows with distance and shrinks with size. A main action tucked in a corner, or tiny and far from where the thumb or cursor is, slows every use.

Applies to: Primary and frequent actions.

BEH-18

MinorAttention

The one important thing stands out (isolation effect)

People remember and notice what differs from its surroundings. When five things are highlighted, nothing is.

Applies to: Screens with one thing that matters most.

BEH-19

MinorAttention

Key items sit first or last (serial position effect)

People remember and choose the first and last items best; the middle gets skipped. Bury the important option in the middle and it's missed.

Applies to: Lists, menus and navigation.

BEH-20

MinorLoad

Long information is chunked

People hold about four chunks at once. '4829 1173 0062' is read and checked; '482911730062' is mistyped.

Applies to: Long numbers, codes, IBANs, phone numbers and long lists.

BEH-21

MinorLoad

Familiar conventions, unless breaking them clearly helps (Jakob's law)

People spend most of their time in other products and expect yours to work the same. A novel pattern costs learning on every first use.

Applies to: Icons, navigation, controls and their placement.

BEH-22

MinorLoad

The product absorbs the complexity, not the user (Tesler's law)

Some complexity can't be removed, only moved. Making people format a date, pick a timezone or know a code shifts the work and the errors to them.

Applies to: Inputs, formatting and setup.

BEH-23

MinorAttention

It feels fast: instant feedback, then progress (Doherty threshold)

Responses under ~400 ms keep people in flow; slower ones need immediate acknowledgement, skeletons or optimistic updates, or people tap again and leave.

Applies to: Actions that take time: search, save, AI answers.

BEH-24

MinorMotivation

Unfinished work is visible and easy to resume (Zeigarnik effect)

People remember unfinished tasks and want to close them. Showing what's left and where to pick up turns that pull into completion, instead of starting over.

Applies to: Long tasks, drafts and multi-step setup.

BEH-25

MinorMotivation

A small first step before a big ask (commitment and consistency)

People who take a small step tend to continue. Asking for everything at once, before any commitment, loses most of them.

Applies to: Onboarding and big asks.

BEH-26

MinorMotivation

Value comes before the ask (reciprocity)

People give back after they receive. A sign-up or data request before they've seen any value feels like a toll and gets abandoned or faked.

Applies to: Sign-up walls, permission and data requests.

BEH-27

MajorTrust

Authority signals are real and relevant

People defer to authority cues. A badge, title or endorsement that isn't real or doesn't apply borrows trust it hasn't earned.

Applies to: Expert badges, certifications, 'recommended by'.

BEH-28

MinorChoice

Switching is made easy (status quo bias)

People stick with what they have even when the new option is better. Imports, defaults carried over and a clear 'what changes' lower the cost of moving.

Applies to: Migrations, new versions and switching from a competitor.

BEH-29

MinorAttention

Things look as interactive as they are (affordances and signifiers)

Flat text that's secretly a button gets missed; a card that looks clickable but isn't gets tapped in frustration.

Applies to: Buttons, links, cards and anything clickable or static.

BEH-30

MinorLoad

Same look, same behaviour (Gestalt similarity)

People assume things that look alike work alike. A blue word that's a link in one place and a label in another breaks that, and they stop trusting the visual language.

Applies to: Repeated elements: links, buttons, tags, statuses.

BEH-31

MinorAttention

Built for scanning: the key words come first

People scan in F and Z patterns and read the first words of lines and headings. Front-load what matters; a heading that starts with 'In order to' hides it.

Applies to: Headings, list items and text people scan.

BEH-32

MinorAttention

Important information doesn't look like an ad (banner blindness)

People ignore anything shaped and coloured like a banner. A real warning in an ad-like box is skipped.

Applies to: Important notices, warnings and offers.

BEH-33

MinorAttention

Changes are shown where people are looking (change blindness)

People miss changes outside their focus. A total that updates at the top while they're at the bottom, or an error far from the field, goes unseen.

Applies to: Totals, errors and content that changes after an action.

BEH-34

MinorMotivation

The immediate benefit is shown (present bias)

People overvalue now and discount later. 'Protect your account' loses to 'Skip'; showing what they get today, or making it one tap, gets it done.

Applies to: Security setup, backups, savings, health and other long-term actions.

BEH-35

MinorTrust

Effort estimates are honest (planning fallacy)

People underestimate time, and products do too. An estimate that's wrong by a lot breaks trust at the moment people are committing.

Applies to: Time and effort estimates: 'takes 2 minutes', step counts.

BEH-36

MinorLoad

No insider knowledge needed (curse of knowledge)

The team knows what 'workspace', 'sync' or an acronym means; new users don't. Copy written from the inside leaves them guessing.

Applies to: Instructions, labels and empty states.

BEH-37

MinorFraming

Errors are calm and don't blame (negativity bias)

Bad experiences weigh several times more than good ones. A harsh, blaming or alarming error colours the whole product; a calm one with a way forward doesn't.

Applies to: Errors, warnings and failure messages.

BEH-38

MajorTrust

Showing the work is honest (labour illusion)

Seeing the steps being done makes waits feel worth it and raises trust. Fake steps or theatrical delays are deception that people eventually notice.

Applies to: Waits for search, AI answers and processing.

BEH-39

MinorChoice

Friction is added where a mistake is expensive (positive friction)

Smooth is not always good. A typed confirmation, a review step or a short delay before an irreversible action prevents the errors people regret most.

Applies to: Irreversible, costly or shared actions.

BEH-40

MinorTrust

Permission is asked in context, with the reason (just-in-time)

Asked at launch with no context, people refuse and rarely reconsider. Asked when the feature needs it, with what they get, most agree.

Applies to: Requests for location, camera, contacts, notifications and other permissions.

BEH-41

MinorAttention

Alerts are reserved for what matters (alert fatigue)

Every badge competes for attention; when everything is urgent, people learn to ignore all of it, including what's important.

Applies to: Notifications, badges and red dots.

BEH-42

MinorLoad

Plain language: short sentences, common words (processing fluency)

Easy-to-read text feels more true and trustworthy, and is understood by more people. Long, abstract or legalistic copy is skipped or misread.

Applies to: All copy.

BEH-43

MinorMotivation

People keep control of what's done for them (autonomy)

People resist what they don't choose. A way to see why, adjust or turn off a recommendation keeps them using it; forcing it drives them away.

Applies to: Personalisation, recommendations and AI features.

BEH-44

MinorMotivation

Feedback builds a sense of competence

People stay with what makes them feel capable. Specific, earned feedback on progress motivates; generic praise or constant streak pressure doesn't.

Applies to: Progress, achievements and feedback on skill.

BEH-45

MinorChoice

Undo beats 'Are you sure?' (confirmation fatigue)

People click through confirmations on autopilot, so the dialog doesn't prevent the mistake. An undo after the fact actually does.

Applies to: Confirmation dialogs.

BEH-46

MajorDark patterns

Loss warnings are honest and specific

'You'll lose everything' to stop a cancellation exploits loss aversion and sunk cost. Say exactly what is kept, what goes and when.

Applies to: Cancellation, downgrade and deletion flows.

BEH-47

MajorDark patterns

No nagging after a 'no'

Asking again on every visit after someone declined wears them down rather than persuading them; it's a recognised dark pattern.

Applies to: Prompts, popups and upsells people already declined.

BEH-48

BlockerDark patterns

No disguised ads or bait and switch

Ads made to look like content, or a button that does something other than it says, deceive people into actions they didn't choose.

Applies to: Ads, sponsored content and buttons.

BEH-49

BlockerDark patterns

No forced action or privacy zuckering

Requiring an account, contacts or extra data to do something that doesn't need it, or defaulting to sharing more, takes what people didn't agree to give.

Applies to: Sign-ups, sharing and data collection.

BEH-50

BlockerDark patterns

No trick questions or double negatives

'Uncheck this box if you don't want to not receive offers' is built to be misread, and consent obtained that way isn't consent.

Applies to: Opt-in and opt-out checkboxes and toggles.

10 rules · the full guide

RTL

Right-to-left screens: Hebrew, Arabic, Persian. On by default when the screen has right-to-left text.

RTL-01

MajorLayout

The layout mirrors: navigation, back and reading order start on the right

A left-to-right layout with right-to-left text makes people scan the wrong way and look for Back in the wrong corner.

Applies to: Screens in a right-to-left language.

RTL-02

MinorLayout

Directional icons point the reading direction

A Next arrow pointing right on a right-to-left screen points backwards. Icons with no direction (play, check, clock) stay as they are.

Applies to: Arrows, chevrons, progress bars, sliders and steppers on right-to-left screens.

RTL-03

MajorText

Mixed text keeps its order

Without correct direction, a brand name, a version number or a final period jumps to the wrong end and the sentence reads scrambled.

Applies to: Right-to-left text that contains Latin words, numbers, emails, links or punctuation.

RTL-04

MajorText

Numbers and currency read correctly

Digits stay left to right inside right-to-left text; a currency sign on the wrong side or a reversed phone number makes the value wrong or untrustworthy.

Applies to: Prices, amounts, phone and card numbers on right-to-left screens.

RTL-05

MinorText

Text aligns to the right edge

Right-to-left text aligned left leaves a ragged starting edge and reads as an unfinished translation.

Applies to: Paragraphs, labels and lists in a right-to-left language.

RTL-06

MinorForms

Fields type in the direction of their content

Name and address fields type right to left; email, URL, phone and code fields type left to right. The wrong direction scrambles what people enter.

Applies to: Inputs on right-to-left screens.

RTL-07

MinorLanguage

No leftover strings in another language

An English 'Cancel' or error message on a Hebrew screen reads as unfinished, and some people won't understand it.

Applies to: Buttons, labels, errors and empty states on a right-to-left screen.

RTL-08

MinorLanguage

Dates, times and units use the local format

03/04 means 3 April in Israel and 4 March in the US; the wrong format causes missed appointments and wrong orders.

Applies to: Dates, times and units on right-to-left screens.

RTL-09

MinorTypography

The type is set for the script

Letter-spacing, all-caps styles, tight line height or a font without the script's glyphs breaks Hebrew and Arabic text, and Arabic letters must stay joined.

Applies to: Hebrew, Arabic and Persian text.

RTL-10

MajorAI

AI messages follow the direction of their own text

An English answer forced right to left, or a Hebrew answer forced left to right, is hard to read. In a right-to-left chat the user's bubbles sit on the left.

Applies to: Chats and AI output in a right-to-left product.

In Figma

Run the checklist where you design.

Crit Gate runs the same rules on your Figma frames, pins findings on the canvas and fixes them for you.

↑