An attractive generated screen can hide a weak product decision. Before adjusting colors, ask what the design enables someone to do and whether the interaction matches their expectations.
Review the task first
Name the user, goal, starting context, and success condition. Walk through the complete task. Can someone tell where they are, what happens next, and whether their action worked? Look for steps that add effort without reducing uncertainty.
Try the uncomfortable states
Review empty results, invalid input, interrupted connections, long content, expired sessions, and destructive actions. A polished happy path is only one part of a usable product. Check that recovery preserves useful work and explains what to do next.
Inspect accessibility and content
Try the keyboard, visible focus, text enlargement, labels, and meaningful reading order. Do not use color alone to communicate an error. Check contrast with a suitable tool and evaluate the implementation with assistive technology where relevant. Automated checks can help, but cannot establish accessibility on their own.
Challenge the logic
Generated interfaces sometimes introduce unsupported features, mismatched labels, or controls that appear interactive but do nothing. Trace each control to an intended behavior. Delete features that do not serve the task.
Turn feedback into a decision
“This feels wrong” is difficult to act on. Try “The success message disappears before the user can confirm the appointment details.” Describe the problem, likely consequence, and next action. Rank issues by impact and confidence rather than personal taste.
Further reading: W3C WAI: Introduction to Web Accessibility.
Try it yourself
Review one generated screen using four headings: task, states, accessibility, and logic. Identify the three highest-impact issues and one experiment to check your assumptions.