24

Every dashboard I one-shotted came out as the identical layout: three stat cards on top, a big line chart, a table. Technically fine, visually indistinguishable from every other AI dashboard on the internet.

The change that mattered was giving the model a real information hierarchy brief (what the primary decision is, what's secondary) and a hard ban on the cliches, plus a density target. It stopped padding everything to 24px and started grouping related metrics.

Curious how others force genuine layout variety without hand-designing every screen.

THE PROMPT
Design and build a React + Tailwind operations dashboard for {DOMAIN, e.g. a support desk}. Before coding, write a 4-line brief: the single primary decision this screen supports, the 2 secondary questions, and what can be hidden behind a click.

Layout rules (enforced):
- The primary metric gets a visually dominant zone (roughly 40% of the top fold), not an equal-sized card in a row of four.
- Group related numbers into labeled clusters; do NOT emit a flat row of identical stat cards.
- Max one hero chart above the fold; everything else earns its place.
- Information density: aim for a 12-16px base spacing scale, not the default 24px everywhere. Tables use 32px rows, not 56px.

Banned: three-equal-cards-then-a-line-chart, generic 'Total Users / Revenue / Active' placeholders, drop shadows on every element, gray-on-gray with no accent, pie charts. Use exactly one accent color plus semantic red/amber/green for status.

Accessibility: every chart has an aria-label summarizing its takeaway; status is never encoded by color alone (add an icon or text). Ship it as one file with realistic sample data that tells a small story (one metric is trending badly).
Banning drop shadows on every element instantly de-slops the output. The uniform shadow is the tell.designbynight 1 month ago
32px table rows vs 56px is such a small lever with a huge effect on how 'pro' it feels. Adding this to my base prompt.css_cyra 1 month ago
add a comment

1 Answer

20

The 'write the brief before coding' step is what breaks the template reflex. I extend it with a forced constraint: name one thing the dashboard deliberately does NOT show, and one metric that gets an annotation explaining why it's bad right now. Annotations are what make it read as designed by a human who cares.

THE PROMPT
Add to the brief: (1) one metric intentionally omitted and why, (2) one on-chart annotation calling out the current anomaly in plain language ('deploys up 3x since Tuesday, likely the cause'). Render annotations as a labeled callout with a leader line, not a tooltip.

Your Answer