Skip to content
uxdesign
coach.
Get a UX review

THE AI FLOW REVIEW KIT / FREE TO USE

Review your AI flow.
Know what to fix first.

Turn “this looks fine” into three improvements you can explain—and a plan to test them.

For solo UX designers and small product teams reviewing an AI-generated booking, signup or onboarding flow before a client or stakeholder handoff.

Free · No signup needed · Allow about an hour for your first review.

WHAT YOU WILL MAKE

Three priorities. A clear next test.

  1. A defined taskBook a consultation at the intended time.
  2. Three grounded prioritiesShow the timezone. Clarify confirmation. Preserve form data.
  3. A useful next testCan someone book, recover and explain what happened?
FICTIONAL PRACTICE EXAMPLE

SEE THE RESULT FIRST

A polished booking screen can still leave people guessing.

Imagine an AI-generated consultation flow with a time picker, contact form and “Done” message. The visual design looks complete. The review asks whether the person can finish the task with confidence.

Fictional booking example. Before: 10:00 without a timezone, a Submit button and Done message. Proposed: 10:00 AM Eastern, Confirm consultation, and persistent appointment details. These are proposed improvements, not measured results.
Illustrative wireframes, not a live booking interface. Proposed changes still need implementation and testing.
What the finished review says
Observation / sourceRisk to investigateDecision / next check
1 · Timezone missing
Illustrative screen A shows “10:00” with no timezone.
A visitor elsewhere may expect the wrong appointment time. No participant evidence yet.Display the selected timezone alongside available times and confirmation. Ask someone in another timezone to explain their booking time.
2 · Confirmation is vague
Illustrative screen C only says “Done.”
The person may be unsure whether a booking exists or what happens next.Keep date, time, timezone and next step visible after success. Test whether a participant can accurately explain the outcome.
3 · Recovery is unspecified
The supplied mockups contain no network-error state.
Entered details could be lost if submission fails. This is an unknown, not an observed defect.Specify a retry state that preserves inputs. Simulate failure in the working prototype and check for duplicate bookings.

ONE FLOW / SIX SMALL STEPS

Review the flow. Decide the next step.

Bring one flow with up to five screens, the intended user goal, known constraints and any evidence you have permission to use. A screenshot supports visual inspection; a working prototype is needed to check behavior.

Step 1: frame the task with a user, goal and success condition.01 / FRAME · 5 MIN

Name the task, not the screen.

Write who is acting, where they start and what success means. Example: “A new customer chooses an available consultation and can repeat the confirmed date, time and timezone.”

Output: one task statement and the constraints. Record unknowns rather than filling them in.

Step 2: trace the start, action, confirmation and recovery states.02 / TRACE · 10 MIN

Walk the whole journey.

Number the screens A–E. Follow the main path, back navigation, empty results, invalid input and an interrupted connection. Record a screenshot label or prototype step for every observation.

Output: a small state inventory. “Not supplied” means unknown, not broken.

Step 3: compare an observation, a possible consequence and evidence still needed.03 / QUESTION · 10 MIN

Ask AI for candidates. Check each one.

Use the prompt below in a tool you already have access to, with approved screenshots or a text description. Compare each suggestion with the source. Remove invented behavior and unsupported claims about users.

Output: a finding with a source, a hypothesis and a verification gap.

Step 4: check keyboard operation, labels, recovery and visible feedback.04 / VERIFY · 10 MIN

Try what the image cannot show.

On a working prototype, check keyboard order, visible focus, labels, text enlargement, error messages and preserved inputs. Measure contrast with a suitable tool. Check that every control has the intended result.

Output: passed, failed or not tested for each check. These checks do not establish accessibility conformance.

Step 5: prioritize blockers first, then confusion and polish.05 / DECIDE · 10 MIN

Choose three improvements.

Fix a blocked task, lost work or misleading outcome before visual polish. For each priority, state impact, confidence, effort and a reason. If evidence is weak, choose a test before committing to a large redesign.

Output: three decisions, not a long AI-generated scorecard.

Step 6: change one thing, test with people and record a decision.06 / PREPARE · 10 MIN

Make the next test concrete.

Choose a change, a realistic participant task and an observation that would challenge your recommendation. “Book a consultation for next Tuesday morning” is more useful than “Do you like this design?”

Output: a test plan with an owner. Run it with appropriate participants afterward; AI feedback is not user research.

Copy the critique prompt
Act as a UX critique partner. Review only the material I provide.
User and task: [who, goal, starting context]
Success condition: [observable outcome]
Screens / prototype: [A–E, descriptions or approved attachments]
Evidence: [real observations with source IDs, or “none yet”]
Constraints: [technical, business and accessibility needs]

Return up to five candidate issues. For each, give:
1. The observable detail and exact source/screen.
2. The potential consequence, explicitly labeled a hypothesis.
3. Whether behavior is observed, inferred or unknown.
4. A proposed change and the tradeoff it introduces.
5. A practical check that could disprove the concern.

Include empty, error, loading, success and recovery states. Mark states you cannot inspect as not tested. Do not invent interviews, quotes, conversion results, accessibility scores or features. End with three recommended priorities and the evidence missing for each. A human will verify every recommendation.

Use anonymized material you are authorized to share. Check your organization’s AI policy before uploading client work.

MAKE IT YOURS / FREE EDITABLE WORKSPACE

Start with an example.
Leave with your own review.

Edit the task, findings and test plan below. Drafts stay in this browser on this device; they are not sent to Ashley. Export a copy before you leave or clear browser storage. Analytics records button actions only, not your entries.

Loading the example…

Your three priorities

Keep observed details separate from predicted consequences. Include a source or mark the evidence gap.

Markdown opens in a text editor and can be pasted into a document. JSON can be reimported here. Download a blank template without using the editor.

WANT AN EXPERIENCED SECOND OPINION?

Review one flow
with Ashley.

Bring your AI-generated prototype and the draft you made here. The UX Decision Review helps you decide what to change, why it matters and what to test next.

$750 fixed scope
Discuss my flow · free 15-minute call →

The fit call is free. Scope, timing and payment are agreed before paid work begins. Your worksheet is not automatically shared.

See the full review scope →

One journey. Three priorities.

  • Review of one journey, up to five screens
  • Annotated findings with rationale
  • Three prioritized improvements
  • 60-minute working review
  • Practical test plan and one clarification round

Bring: a viewable prototype or screenshots, the intended user task, constraints and any existing evidence. Use sanitized examples until an appropriate sharing method is agreed.

This is a focused expert review. A full redesign, implementation, participant research and accessibility certification require separate scope.

Use the kit when the next decision is unclear.

“The AI gave me a beautiful screen. Is it usable?”

Start with the task and states. A screenshot cannot show keyboard behavior, server errors or whether the last action actually succeeded. Record those as not tested and investigate them in a working prototype.

“My stakeholder wants to know why we should change it.”

Export the review. Show the exact observation, the consequence you are concerned about and how you will check it. Keep the original and proposed states side by side. Explain the tradeoff, such as adding detail for clarity at the cost of a little screen space.

“I have no research yet. Can I still use this?”

Yes, for an initial expert critique. Say that the recommendations are hypotheses. Use the test plan to gather real evidence; do not create synthetic testimonials or present an AI prediction as a participant finding.

“Do I need a paid AI subscription or Figma?”

No special tool is required for the worksheet. Use screenshots, descriptions or a prototype you already have. AI is optional. Tool capabilities and access vary; this guide does not depend on a particular model or integration.

“Will my draft be sent to the coach?”

No. It is stored locally in this browser and exported only when you choose. Booking a fit call does not upload it. Avoid storing confidential material on a shared device, and use the clear button when finished.

Keep building your judgment.

Sources and further reading

This original workflow applies a bounded review to a fictional booking example. Related guidance supports defining evaluation criteria, reviewing AI-generated output and checking accessibility beyond the screenshot.

Expanded November 11, 2026 · UX Design Coach · Fictional example; no measured results claimed.