Iterating via Conversation
Chat is the primary way to steer Claude Design. You're not prompting once and done, you're having a design review, in writing, with an AI that can make changes in seconds. The patterns that work in a real design critique work here too.
Here's an iteration pass on Lineup, asking Claude to add a sidebar to the kanban board we generated in the previous lesson. Watch how Claude reads the existing canvas, proposes the sidebar structure, and drops it in without disturbing the column layout.
What chat is best for
- Broad changes: 'Make the color scheme warmer' or 'Swap the hero and pricing sections'
- Structural moves: 'Add a testimonial block between features and footer'
- Alternative directions: 'Show me three layouts for the hero, centered, split, and full-bleed'
- Content-level edits: 'Rewrite the headline to emphasize speed instead of accuracy'
What chat is slower for
- Changing one specific element (use inline comments, next lesson)
- Micro-adjustments to spacing or color values (direct canvas edits)
- Producing variations (ask for sliders, lesson 6)
How to structure iteration messages
State what's working
Tell Claude what to keep. 'The hero layout is right; keep the three-column feature grid.' This prevents Claude from re-doing things you liked.
State what's not working
Be specific. 'The pricing cards feel cramped, too much text per card.' Much better than 'pricing looks bad.'
State what to change it to
Give direction, not just a problem. 'Reduce pricing cards to three bullet points each and use a lighter-weight body font.'
Claude Design uses Claude Opus 4.7, Anthropic's most capable vision model. It reasons about your current canvas while processing your message, so references like 'the second feature card' work without extra context.
You can iterate via chat
- Know what to ask chat vs. other tools
- Can frame changes with keep-fix-direction
- Aware of Opus 4.7 vision context
Next: targeting specific canvas elements with inline comments.