Skip to content

Before / After: Real prompt upgrades

Three real patterns. Same intent. Different output quality.


Weak prompt:

Add an accordion.

What you get: A generic accordion, often with the wrong groupings, sometimes a new component when you already had one.

Better prompt:

This settings page feels too dense. Recommend whether accordion sections, tabs, a right-side drawer, or progressive disclosure would make it easier to use. Explain the tradeoffs, then implement the best option.

What you get: A diagnosis, three pattern options with tradeoffs, a recommendation, and an implementation that uses your existing primitives.

Why the second works:

  • Names the outcome (“feels too dense”) instead of the component.
  • Lists rival patterns so Claude has to differentiate.
  • Asks for the tradeoff before the implementation.

Weak prompt:

Add a menu.

What you get: A dropdown menu with whatever actions Claude guesses are relevant. Half the time it’s a <select> — not a menu at all.

Better prompt:

This page has too many visible actions. Recommend the best action organization pattern: dropdown menu, toolbar, split button, context menu, or command palette. Explain which actions should stay visible and which should move into the menu.

What you get: A pattern recommendation, an audit of which actions are primary vs. secondary, and an implementation that respects the difference.

Why the second works:

  • Lists the rival patterns by name — so you learn the words.
  • Asks Claude to decide visibility for each action, not just dump them in a menu.
  • Acknowledges the real problem: too many visible actions, not no menu.

Weak prompt:

Make the dashboard better.

What you get: Padding tweaks, a slight font change, maybe a new color. You ship and it still feels like a prototype.

Better prompt:

Improve this dashboard for a user who needs to understand performance quickly and take action. Focus on metric hierarchy, filters, drill-downs, table actions, empty states, loading states, and responsive layout. Give me 3 design options before coding.

What you get: A real product proposal. Probably KPI cards with trend indicators, a date range picker, faceted filters, a drill-down side panel, skeleton loaders. Three of them, with tradeoffs.

Why the second works:

  • Names a target user (“needs to understand performance quickly and take action”).
  • Lists specific concerns (hierarchy, filters, drill-downs, states, responsive).
  • Demands 3 options.

Weak prompts sayBetter prompts say
Add a componentRecommend a pattern
Make it betterRecommend an outcome for a specific user
Use a thingCompare 3 candidate patterns and pick
Just do itExplain tradeoffs, then implement

The trick: describe the user problem and the rival patterns, not the component.

Claude is a much better designer when you give it a design problem. It’s a mediocre designer when you give it a component request.