19 Best Claude Prompts for Design That Actually Work

Design work stalls far more often on the conceptual side than the technical one — a vague brief, an idea you cannot articulate, a client who needs the reasoning spelled out. That is exactly where Claude prompts for design earn their place. Claude does not produce visuals, but it produces the thinking that has to happen before and around them.

Below are 19 prompts covering briefing, type and colour systems, interface copy, critique, and handoff. Each is ready to paste. Run them in Chat Smith, a free AI chat that keeps Claude, GPT and Gemini in one place, and see the wider set of Claude prompts for work outside design.

Claude cannot see your file. Everything it knows about your design comes from how precisely you describe it, which means the quality of the output tracks the quality of your description almost exactly. A long-context model such as Claude Sonnet 5 can hold a brief, a brand guide and three rounds of critique in one conversation — which is what makes iterative design work possible rather than one-off. Four habits separate a useful prompt from a generic one.

Describe what is on screen, not what you intended. Layout, hierarchy, spacing, colour values, type sizes. A description of your intention gets you feedback on your intention.

Name the audience and the decision. “B2B SaaS buyers evaluating three vendors” produces different advice from “users”. The decision the design has to drive matters more than the aesthetic.

Ask for the reasoning, not just the answer. A font pairing you cannot justify is one you cannot defend in a client meeting. Every recommendation should arrive with its rationale attached.

Verify anything checkable. Font availability, licensing, and contrast ratios are all things a model can state confidently and get wrong. Check them in the type foundry and a contrast tool before they reach a client.

Four prompts for the stage before anything is drawn. Most projects that go wrong went wrong here — and a verbal mood board is far faster to iterate on than a visual one, though a text to image tool is useful once a direction is chosen.

1. The creative brief builder

Write a detailed creative brief for [project]. Client and brand: [describe]. Target audience: [who they are, what they currently use, what they believe]. The brand should feel [three adjectives]. Include brand overview, audience, design objectives, tone of voice, deliverables and constraints. Then flag anything in my input that is contradictory or too vague to design against.

2. The verbal mood board

Describe four genuinely distinct visual directions for [project]. For each: colour palette with hex suggestions, typography personality, imagery style, UI component feel, and the emotion it should produce. Make them different in kind rather than degree — no two should be variations of the same idea. Then tell me which is riskiest and why.

3. The category convention audit

Here is how the main players in [category] present themselves visually: [describe their sites, colours, type, imagery]. Identify the conventions of this category — the things every one of them does. Then tell me which conventions exist for a real usability reason and which are just imitation, since the second group is where differentiation is safe.

4. The audience translator

My audience is [describe them: role, context, what they are doing when they arrive, what they are anxious about]. Translate that into design implications: what they need to see first, what will make them distrust the page, how much text they will actually read, and which device and conditions they are likely using. Be specific enough that I could act on each point.

Four prompts for the systematic layer. Naming and structure decisions here are cheap to change now and expensive to change once developers have built against them.

5. The typography pairing

Suggest three typography pairings for [describe the brand and industry], which should feel [three adjectives]. For each, give the heading font, body font, and the reasoning I could use in a client meeting. Restrict yourself to families you are confident are available on [Google Fonts / Adobe Fonts], and tell me where I should verify licensing myself.

6. The type scale

Build a type scale for [product type] with [number] steps, base size [X]px. Give me the size, line height, weight and intended use for each step, plus the mobile values. Then tell me which two steps are close enough that a designer will misuse them, and whether the scale needs one of them at all.

7. The palette namer

Here is my palette: [list each colour with its hex value]. The brand is [describe]. Give each colour an evocative, memorable name that fits the brand, plus a one-line description of its role in the system. Avoid names that describe the colour itself — they add nothing that the hex does not already say.

8. The token naming convention

I am building a design system for [product]. Generate a semantic colour token naming convention covering brand colours, a neutral scale, status colours and background/foreground pairs. Format as a table. Then explain the rule behind the naming in one paragraph, so a developer joining in six months can extend it without asking me.

Four prompts for the words inside the interface, which shape the experience more than most visual decisions do. Run the final strings through an AI grammar checker before handoff, and keep the tone consistent with whatever your AI ad copy generator is producing for the marketing site.

9. The onboarding microcopy

Write the microcopy for a [number]-screen onboarding flow for [product]. For each screen: headline, body of no more than two sentences, and a button label. Tone: [describe]. Each screen must earn the next one — if a screen only exists to explain a feature, tell me to cut it rather than writing copy for it.

10. The empty and error states

Write the empty states and error messages for [describe the product area]. For each: what happened, what the user can do about it, and the button label. No apologising, no blaming the user, no “oops”. Where an error is our fault, say so plainly. Where the user cannot fix it, tell them what we are doing instead.

11. The form labels and helper text

Here are the fields in my form: [list them with their purpose]. Write the label, placeholder and helper text for each. Labels stay visible rather than living in the placeholder. Then flag any field where the helper text is doing work the label should be doing, and any field I could remove entirely.

12. The button and action audit

Here is every button label in [describe the flow or screen]: [list them]. Check them for consistency of voice and grammatical form, ambiguity about what will happen after the click, and any two labels that could be confused. Rewrite the problematic ones and tell me which single action on this screen should be visually dominant.

Four prompts for the review pass. Describe the design as it exists rather than as you meant it — the whole value of this section depends on that, since a flattering description produces flattering feedback.

13. The layout critique

Here is my design: [describe layout, hierarchy, colour, type, spacing and what is above the fold]. Audience: [describe]. The action I want them to take: [state it]. Critique it against conversion-focused principles and give me five specific improvements ranked by expected impact. Tell me which one I will resist and why it is still right.

14. The hierarchy check

Here is everything on the screen, listed in the order it appears: [list the elements with their size and weight]. Tell me the order a user will actually notice them in, and whether that matches the order of importance I intended: [state your intended order]. Where they diverge, tell me what to change.

15. The accessibility review

Here is my design: [describe colours with hex values, type sizes, interactive elements, and states]. Review it for accessibility: colour contrast, text size, touch target size, reliance on colour alone to convey meaning, focus states, and motion. Tell me what to verify with a real contrast checker and screen reader rather than taking your word for it.

16. The not-working diagnostic

Something is wrong with this design and I cannot name it: [describe it in as much detail as you can]. Do not suggest fixes yet. Ask me up to six questions about intent, constraints and context, then give me your best diagnosis of what is actually wrong. It is often not the thing I am staring at.

Three prompts for the part where good work gets approved or rejected on how well it is explained. For the document itself, an AI proposal generator handles the structure once the argument is settled, and an AI storyboard generator helps when you need to show a flow rather than a screen.

17. The design rationale

Write a three-paragraph design rationale for [describe what you made and the specific decisions: colour, type, composition, imagery]. Audience: a non-designer [founder / client / stakeholder]. Explain each decision in terms of what it does for their business, not in terms of design theory. No jargon that would need explaining.

18. The brand voice guide

Create a brand voice and personality guide for [describe the company and audience]. Include the archetype, four personality traits with descriptions, tone principles, words to use and avoid, and three before-and-after copy rewrites. Then add one section on how the voice changes in error states and support — the part most voice guides omit.

19. The developer handoff notes

Here is my design: [describe the components, states and behaviours]. Write handoff notes for a developer: what is fixed versus fluid, how each component behaves at breakpoints, every interactive state, and what happens with unexpected content — long names, missing images, empty lists. Then list the questions a developer would ask that my description does not answer.

Design work runs the same prompts repeatedly with different projects, which makes them worth saving properly. Chat Smith lets you keep your brief template, your critique prompt and your handoff checklist as one-click templates, group them by client, and share the set with the rest of a product team so briefs arrive in a consistent shape.

It also runs several AI models side by side — Claude, GPT, Gemini, DeepSeek and Grok. That has a practical use in design: models differ noticeably in register, and the one that writes a conceptually rich brief is rarely the one that produces the tightest button labels. Sending the same prompt to two of them takes seconds and usually settles which output you ship.

Start with prompt 16, the not-working diagnostic, on whatever is currently sitting unfinished in your file. It is the only prompt here that asks you questions before giving you an answer, and it is usually the one that finds the actual problem.