The dashboard is on the shared screen, the designer has walked through the new layout, and the first comment is "it feels a bit off." The room nods. Someone suggests more white space, someone else asks about the blue, and the discussion drifts to a competitor's product for eight minutes. The meeting ends with no decision and the same file still open.
That pattern costs more than one meeting. Unresolved critique pushes the work into a fifth and sixth round, where the designer guesses at intent and rebuilds the same screen twice. Reviewers stop trusting their own notes, engineers start asking which version is real, and the redesigned dashboard ships with the same unclear filter placement that started the conversation.
The fix isn't a better meeting or a louder voice. It's the shape of the comment. The design critique examples below each pair a specific critique with the reasoning behind it, the UX change it points toward, and a test you can run to check whether the critique was right. Copy one format, run it on a single screen next week, and see what changes.
1. I Like, I Wish, I Wonder Critique of a Dashboard Redesign
A dashboard review collects every kind of ambiguity at once, which makes it a good place to start.

The format sorts comments into three lanes. "I like" names what works and should be protected. "I wish" names a change with a reason attached. "I wonder" names an open question instead of a demand. It comes from studio teaching rather than one published author, and its value is timing. It fits the review where the team needs direction, not a verdict.
Say a product team is reviewing a redesigned analytics dashboard. One reviewer writes that they like the date filter placement because it survives a resize, wish the export button were visible without scrolling, and wonder whether churn users will find the segment picker in their first week. Three lines, three named elements, no guessing about what to do next.
Where the format goes soft
"I wish" comments rot into preference when they skip the reason. A wish that the header were lighter gives the designer a task but no goal. A useful wish names the user it serves and the moment it affects.
One reviewer's pass also isn't enough. Nielsen and Landauer's six case studies, summarized in the NN/g heuristic evaluation resource, found that a single evaluator surfaced about 34% to 35% of usability problems in an interface, with individual runs ranging from 19% to 51%. Run the format with three reviewers and consolidate, or you're approving a screen on a third of the evidence.
Practical rule: Convert every "I wonder" into a question with an owner and a date, or it becomes the opening line of the next meeting.
- Tie each wish to a user goal. State who is affected and what they're trying to finish before you propose the change.
- Keep the review to one screen. Narrow scope keeps comments specific and stops the conversation from wandering into navigation.
- Test the wonders. A PM who flags the segment picker should book five user sessions to find out whether the worry is real.
For a wider set of starting points, browse real analytics dashboard designs before the session and note which filter patterns your layout already matches.
2. Contrast Failure Critique on a Settings Page
Helper text under a settings toggle looks fine on a calibrated monitor in a bright room, then disappears on a phone in daylight. Reviews miss it because the mockup is viewed on the best screen in the office.
The note that settles the argument
Reviewer note: #9AA0A6 on a white background measures about 2.6 to 1. That fails the minimum for normal text, so move the label to a darker gray token.
The number is calculated, not felt. A light gray such as #9AA0A6 has a relative luminance of roughly 0.347, so the ratio comes out at (1.05) divided by (0.347 + 0.05), or about 2.6 to 1. A designer can rerun that arithmetic in any contrast tool in under a minute, which is what makes the comment land.
The threshold comes from the W3C guidance on contrast minimum. WCAG 2.2 Success Criterion 1.4.3 requires a ratio of at least 4.5 to 1 for normal text and 3 to 1 for large-scale text, while Criterion 1.4.11 sets 3 to 1 for the visual information that identifies controls and focus indicators. Inactive controls, decorative text, and logos are excepted, so the critique should record why an apparent exception applies rather than waving the whole screen through.
Why the pedantry pays
Accessibility defects behave like operational risk, and they respond to operational measures. A healthcare accessibility redesign case study paired the technical score with service metrics: digital task completion rose from 62% to 91% and UI-related support tickets fell from 680 to 120 per month, alongside a Lighthouse score that moved from 49 to 98.
- Measure during the review. Open a contrast tool with the file, not after handoff, so the fix happens before the token reaches code.
- Log the ratio and the criterion. A comment with a number and a success criterion survives a stakeholder debate that a comment about "feeling faint" doesn't.
- Check the exempt states anyway. Disabled labels and placeholders are where teams assume an exception applies without verifying it.
If you want to see how Figr approaches color accessibility inside a review, the color contrast checker walkthrough shows the check in context.
3. Empty State Critique on a First-Run Experience
A new user's first minute inside a reporting tool is usually an empty screen. Many reviews never see it, because the file on screen is full of populated charts.
Start where the user starts
Take a hypothetical empty report screen: a gray panel, a small icon, no text. The critique asks three questions. Does the screen say what this area is for? Does it name the next action? Does it hint at what the finished result looks like?
The third question maps to Jakob Nielsen's visibility of system status heuristic, which holds that a system should always keep users informed about what's happening. An unexplained gray panel doesn't tell a first-timer whether the report is loading, broken, or empty.
The revision is unglamorous and fast. A one-sentence explanation, a "Create first report" button above the fold, and a faded sample chart with three labeled data points give the user something to recognize and a place to go.
Run the state as a stranger
Most empty-state arguments dissolve once a reviewer narrates the screen out loud as someone who has never used the product. That narration usually surfaces two or three missing sentences, and the designer can fix them without touching the logic.
- Review empty, loading, and error together. These three states share the same slot in the layout, and critiquing them in one pass keeps the language consistent.
- Check the primary action without scrolling. On a 375-pixel-wide phone the empty panel often pushes the button below the fold.
- Ask what the user sees after the first action. The empty state ends the moment data exists, so the critique should cover the transition, not just the blank screen.
The pattern generalizes. Any screen that can render with zero records needs its own review, and first-run screens deserve the earliest one.
4. Error Recovery Critique in a Checkout Flow
Generic errors at the top of a form are the most common failure in this category. They tell the user something went wrong, then leave them to hunt for the field that caused it.
Build the failure matrix first
Take a hypothetical checkout with an invalid postal code. The user submits, the form clears every field, and a red line appears at the top saying the details couldn't be processed. The critique asks the designer to preserve the entered address, highlight only the postal code field, and write a message that says what to type.
That maps to Nielsen's heuristic about helping users recognize, diagnose, and recover from errors, and the practical version is a small matrix built before the review. One row per failure case, three columns: the message the user sees, the field state it produces, and the recovery path. A declined card, an expired promo code, and a missing apartment number each get a row.
Check the baseline in the wild
Form-level defects are the norm rather than the exception. WebAIM's 2025 analysis, collected in Figma's web design statistics reference, found detectable WCAG failures on 94.8% of the top one million homepages, with missing form labels on 48.2%. The field-level labeling and messaging you're critiquing is routinely absent elsewhere, which is exactly why it's worth the argument.
- Test each row in a staging build. A prototype that swallows a bad postal code tells you nothing about what production does with it.
- Confirm screen readers announce the error. Color and outline changes are invisible to assistive technology unless the message is wired to the field.
- Name the recovery cost. Clearing four fields after one typo is a reason for abandonment, not a styling nitpick.
For the pattern library behind this, the guide to handling failures gracefully in UX covers inline validation, message tone, and recovery paths in more depth.
5. Design System Token Drift Critique
Three shades of blue appear on a new onboarding screen. None of them exist in the design system, and the designer who chose them has already moved to another project.
Token drift arrives slowly. One off-palette blue appears in a single flow, then spreads into components and handoff specs until the product no longer matches its own library. The critique asks reviewers to compare each style value against the source library and flag anything that doesn't map to a token.
Three untracked blues on one screen is a documentation gap wearing a color problem's clothes. The next designer will hit the same gap and invent a fourth blue.
How drift survives review
Most reviewers look at layout and flow, not at hex values. A screenshot hides token data entirely, so a screen built from hardcoded values looks identical to one built from the library until an engineer inspects the code.
The documentation gap is measurable. In a survey of 250 engineering leaders by Colab Software, 45% of design standards and guidelines were not documented, up to date, or consistently referenced in design reviews in 2025. When the source of truth isn't referenced during review, drift has nowhere to be caught except late.
- Audit tokens before the review starts. Run a style audit on the frame and bring the mismatch count to the session as a fact rather than an opinion.
- Replace, then document the gap. Swap the three blues for the brand primary token and log the missing variant with the system owner so the gap shrinks instead of returning.
- Track drift per release. A count that climbs across two releases means the system is fragmenting faster than it's being maintained.
The harder version of this problem sits between design and code, which is why Figma components and code diverge over time even when the file looks clean.
6. Prototype Fidelity Critique Before Handoff
Imagine a PM reviewing a prototype that shows a six-step checkout while the production app runs four. Nobody flagged it, because the prototype looked polished and everyone assumed it was accurate.
Fidelity critique compares the prototype against the live product and asks which differences are intentional. Generic components, invented navigation, and simplified data all pass a visual review and then fail in development, because engineering can't build the flow the designer drew.
Separate the intended change from the drift
The review produces two lists, not one. The first is planned changes with a reason and a target release. The second is drift to correct in the prototype. The six-step checkout belongs in the first list only if the team intends to add two steps, with a date attached.
This is what I mean by reviewing the prototype as a proposal rather than a picture. A picture gets approved on appearance, while a proposal gets evaluated on feasibility, and feasibility is where handoff conversations break down.
- Capture the live app before the session. Screenshots of the current product give every reviewer the same baseline, so comparisons stop relying on memory.
- Label each difference. The words "intentional change" or "unintended drift" next to each item turn a messy discussion into a decision log.
- Run one task-based test first. A single task on the prototype reveals whether the extra steps affect completion before engineering spends a sprint on them.
The prototype fidelity decision framework is useful when the team disagrees about how faithful the artifact needs to be for the question at hand.
7. Goal-Based Critique Versus Taste-Based Critique
An executive asks for the pricing page to "pop more." That is a taste statement in goal-shaped clothing, and it will generate three revisions nobody can evaluate.
The reframe is to name the user task, the audience, and the success condition before anyone judges the layout. "Help a first-time visitor identify the right plan within one screen" is testable. "Make it pop" isn't, and the design that follows it usually fails twice: once in review, once in market.
Replace the verdict with an observation
The useful comment names a context and a perceived value the designer can preserve or test. "This layout works well for a user who compares plans weekly" gives the designer something to check against real behavior. "I like it" gives them nothing to act on. On a dashboard, "the persistent filter makes weekly cohort comparisons possible without reopening a menu" beats "this feels better," because the first version points at a task a test can confirm.
The same discipline applies when methods disagree. Expert review and user testing surface different kinds of problems: reviewers catch consistency issues and pattern violations, while sessions expose where actual users stall. A disagreement about taste is best settled by whichever method matches the question under debate.
Goal critique has a failure mode of its own. A goal worded broadly enough, like "improve usability," approves every layout put in front of it, so the session drifts back to preference with better vocabulary.
A taste argument costs a week of revisions. A goal argument costs one user test.
- Ask for the user task first. Every reviewer states who they're imagining and what that person is trying to do before offering an opinion.
- Write the goal at the top of the review doc. One sentence per screen keeps the session anchored when a senior voice shifts the topic.
- Settle disputes with evidence, not seniority. Analytics or a five-person test decides faster than another round of opinions.
8. A/B Variant Critique Before Launching an Experiment
Two checkout button labels go to review. The conversation stays on wording until someone asks whether the variants differ in phrasing alone or also change the perceived cost of the purchase.
That question is the whole point of variant critique. Each option needs a rationale, the change has to be large enough to register, and the test has to be able to detect a difference at all. Without those three conditions, a team runs an experiment for three weeks and learns nothing it can act on.
What the variant review has to force
A Microsoft Clarity case study of an e-commerce redesign shows the productive version. The team analyzed click maps, scroll maps, and customer behavior, found that users weren't seeing the price-range filter, and repositioned it. About 90% of users saw the filter after the change, and overall conversion rose 12.5%, with 16% on mobile and 7% on desktop.
The split matters more than the headline. The mobile lift was more than double the desktop lift, which points at responsive hierarchy and control visibility rather than visual polish. A variant review that didn't segment by device would have missed the reason the change worked.
- Write the hypothesis before the visuals. One sentence on the user behavior you expect to change, plus the primary metric you'll read.
- Change one meaningful variable. Two labels are comparable. A new label plus a new layout plus a new price display is three experiments wearing a trench coat.
- Ask an analyst about sample size and duration. A test that ends when the result looks good isn't a test.
How the eight design critique examples differ in use, evidence, and limits
The eight examples split into two working groups. Contrast checks, token audits, and fidelity reviews produce findings a reviewer can verify in the same session: a ratio, a token name, a difference against production. The rest produce judgments that depend on outside evidence, whether that is a heuristic, a task test, or a live experiment.
| Item | Best for | Evidence basis | Main limitation |
|---|---|---|---|
| I Like, I Wish, I Wonder Critique of a Dashboard Redesign | Balanced, early-stage design feedback | Three-part feedback format used in design studios and reviews | Can produce shallow feedback and does not prioritize changes |
| Contrast Failure Critique on a Settings Page | Identifying and correcting text contrast failures | WCAG 2.2 contrast thresholds and a calculated example | Checks color contrast only, not broader accessibility |
| Empty State Critique on a First-Run Experience | Improving guidance and next steps for new users | Nielsen's visibility of system status heuristic | Needs real-user observation to validate effectively |
| Error Recovery Critique in a Checkout Flow | Reviewing checkout errors and recovery paths | Nielsen's error recognition, diagnosis, and recovery heuristic | Many failure paths can make reviews lengthy |
| Design System Token Drift Critique | Maintaining consistency with design system tokens | Comparison against a named token set and library | Depends on an up-to-date token library |
| Prototype Fidelity Critique Before Handoff | Checking prototype accuracy before testing or development | Comparison with the current production UI | Exact matching can slow exploration of new ideas |
| Goal-Based Critique Versus Taste-Based Critique | Turning subjective feedback into testable goals | Task-based evaluation and user-goal principles | Brand and craft decisions may resist task-based measures |
| A/B Variant Critique Before Launching an Experiment | Preparing clear, measurable design experiments | Experiment rationale, variable isolation, and outcome criteria | Assessing sample size and statistical power requires analytical skill |
Read the limitation column as a scope warning rather than a defect. A contrast check that only measures color has still done its job; the failure is treating it as an accessibility audit. Breadth and strength of evidence pull in opposite directions across the table, and the narrow formats are often the ones that return a pass or fail answer before the meeting ends.
Choose One Critique Format and Run It on Your Next Screen
The failure mode in many teams isn't a shortage of opinions. It's a shortage of decisions. Feedback arrives, gets discussed, and then evaporates, which is why so many product orgs can't say what changed after a review.
The scale of that loss is visible in the numbers. Colab Software's research found that 43% of design-review feedback was never tracked or addressed in 2025. Nearly half of the critique a team generates has no path to implementation, which makes the format of the review as important as the talent in the room.
In short, pick the format that matches your weakest link rather than the one that sounds most impressive.
If your reviews end in vague discomfort, start with the like, wish, wonder structure and add two more reviewers. If accessibility gets skipped until the audit, start with the contrast note and the error matrix, because those two produce a measurable finding in under ten minutes. If components keep drifting, run the token audit before the session and bring the mismatch count as your opening slide.
Then do the part many teams skip. Write down the decision the review produced, the owner, and the check that will confirm it after release. A critique that ends with a named change and a date has done its job. A critique that ends with agreement has not.
One screen, one format, one week. Run the contrast check on the settings page nobody has touched in a year, or convert a single "it feels off" into a testable question before Friday. If your team uses Figr, you can compare a new screen against your imported design system tokens and components before the review starts, so the drift conversation happens with evidence on screen.
Design critique gets expensive when the reasoning stays in someone's head. Figr reviews an existing screen against UX patterns, accessibility checks, and your own components and tokens, then shows the reasoning behind each recommendation so your team can accept, question, or reject it with something concrete to point at. Capture your live product, import your design system, and bring the output into your next review as a second opinion rather than a verdict. Visit Figr to see how it fits your review process.
