Most designers who have tried Claude and moved on used it the same way they use a search engine: type a question, skim the answer, close the tab. That's a bit like using a bicycle as a coat rack — technically possible, entirely missing the point.
Claude is most valuable not as a lookup tool but as a thinking partner. The difference is in how you engage with it. A question gets you a generic answer. A context-rich prompt with a clear goal and defined constraints gets you something that can genuinely accelerate the work.
This is a guide to how we actually use Claude at Desyn — the specific prompts, the use cases that work, and the ones that don't.
Why Claude, Not ChatGPT
This will ruffle some feathers, but it's based on consistent experience across hundreds of real tasks: Claude is better for the kind of work designers do.
ChatGPT is faster and more confident. It gives you clean, decisive answers quickly. For coding tasks, it's often excellent. But for anything involving nuance — writing that needs to match a specific tone, analysis that requires holding multiple perspectives simultaneously, communication that needs to feel human rather than processed — Claude tends to produce better output.
The difference is most apparent in writing tasks. ChatGPT's copy tends to land in a certain register: polished, slightly corporate, somewhat predictable. Claude's writing feels more considered. It holds back from the obvious move. It asks clarifying questions when the brief is genuinely ambiguous rather than confidently producing something that misses the point.
For designers, who spend a lot of time on writing tasks that need to feel authentic (UX copy, client communications, deck narratives), this distinction matters a lot.
The Brief Breakdown
The first place Claude earns its keep is with client briefs. Most briefs are, to varying degrees, incomplete. The client has a business problem they've articulated as a design request, and there's usually a gap between what they've asked for and what they actually need.
A prompt we use regularly when a new brief comes in:
"Here's a client brief: [paste brief]. You're a senior brand strategist. Identify the three things this brief doesn't say but probably needs to — gaps in the brief where assumptions are being made that could go wrong. Then suggest five questions I should ask the client before starting."
This prompt consistently surfaces things we'd have caught eventually anyway — but catching them before the kickoff call rather than three weeks in is the difference between a smooth project and a pivot conversation.
A variant we use for positioning work: ask Claude to identify what the brief implies about the client's target audience, then check whether that audience matches the product or service being offered. Mismatches here are the source of a lot of confused websites.
Writing Copy with Claude
The most consistently useful application. Claude can write website copy that doesn't sound like it was written by AI — but only if you give it enough to work with.
The mistake most people make is prompting for copy without context. "Write a hero headline for a SaaS product" produces something generic. The prompt that works is longer and more specific:
Prompt template for hero copy: "I'm writing the hero section for [client name], a [what they do] for [who they serve]. Their tone is [adjectives — e.g. 'direct, warm, confident, never jargony']. The key thing visitors need to understand in the first five seconds is [core value proposition]. Write 8 headline options ranging from direct/functional to more conceptual/evocative. Each should be under 10 words."
That prompt generates genuinely usable options roughly 70% of the time. The remaining 30% gives you something to react to — you can paste the output back in and say "none of these feel right, they're too X, try again with more Y."
For body copy, the same principle applies: the more brand context you provide upfront, the less you need to edit the output. A brief paragraph about who the company is, who they serve, and what makes them different will produce copy that's significantly better than a bare request.
Handling Client Feedback
Client feedback emails are a specific genre with specific problems. They mix legitimate design feedback with personal preferences, business concerns with aesthetic opinions, and — most frustratingly — they sometimes say the opposite of what they mean.
We've started running difficult feedback emails through Claude before responding. The prompt:
"Here's a client feedback email: [paste email]. What are the actual design concerns underneath this feedback? Which points are about personal preference versus genuine UX or business issues? And what's the most professional, clear way to respond that acknowledges their concerns while protecting the strategic decisions we've made?"
This is genuinely one of the most useful applications we've found. Claude is excellent at parsing feedback that's emotionally charged or unclear, and at helping you craft responses that are firm without being defensive.
Generating Design Direction
Before reaching for Midjourney or opening a Figma file, we sometimes use Claude for verbal articulation of a visual direction. This sounds abstract, but it's practical: having a clear verbal description of what you're aiming for makes every subsequent visual decision faster.
A prompt that works well: "I'm designing a website for [client]. Their brand values are [list]. Their competitors look like [describe]. We want to differentiate by feeling [describe]. Write a paragraph describing what this website should feel like to visit — not what it looks like, but what experience it creates. Then describe the visual language (colours, typography feeling, layout principles) that would produce that experience."
The output becomes a reference you can share with collaborators, use as a Midjourney prompt foundation, or present to clients as a framing device before you show visuals.
- Use Claude to develop colour palette rationales — why these specific colours for this brand
- Ask it to write the narrative for a design presentation before you build the deck
- Use it to generate microcopy options: button labels, error messages, empty states
- Ask it to critique a sitemap from a user journey perspective
- Use it to research competitors quickly — give it three URLs and ask for a positioning map
Claude's Limitations
It's not all frictionless. There are things Claude handles poorly that are worth knowing before you invest time in a dead end.
Claude doesn't retain memory between sessions. Every conversation starts fresh, which means you need to re-establish context each time. We keep a "context block" for each active project — a 200-word summary of the client, the brief, and the key decisions made — that we paste at the start of any Claude session related to that project.
It also has a tendency to hedge. When you ask for an opinion, it sometimes gives you a balanced view when what you wanted was a decisive take. You can push past this by explicitly asking for a strong recommendation: "I don't need both sides — just tell me which approach you'd choose and why."
Finally, Claude's visual imagination is limited by its inability to see images unless you specifically use a multimodal prompt. For visual critique, you need to describe what you're showing rather than assuming it understands visual context it hasn't been given.
None of these are dealbreakers. They're just the edges of a tool that, within its strengths, is genuinely transformative for how designers work.