Claude Design: how it works, and how to stay under the limit

What Claude Design makes

Four kinds of work. I built the kit around them because each one wants a different brief.

Landing pages and websites. A full page with sections, copy in place and a responsive layout, delivered as HTML you can open in a browser. This is the strongest category and the one where a brand file makes the biggest difference.

Decks and documents. Pitch decks, one-pagers, reports. When speed beats style, Claude routes decks to Claude Slides. When the look matters, you keep the deck in Design and export to PPTX later.

Social and ads. Static posts, ad variants, carousels, at volume. The skill is writing one brief that produces ten variants rather than ten briefs that each produce one.

Prototypes and dashboards. Clickable app screens, admin panels, data views with placeholder numbers. Good enough to put in front of a user before anyone writes production code.

Which Claude plan includes Claude Design

Plan	Claude Design
Pro	Included
Max	Included
Team	Included
Enterprise	Included

Pro, Max, Team and Enterprise all include it. What separates them is how much you can generate in a week, and that is where the trouble starts for people on Pro.

If you are new to Claude altogether, start with my free Claude mastery guide and come back here once chats, projects and artifacts feel familiar.

Why Claude Design burns your weekly limit

In the reviews I read before building this kit, most of a weekly Pro allowance went in 25 to 35 minutes of Design. Half an hour of work, then a wait until the week resets. That number is the reason the kit opens with setup instead of with prompts.

The cause is mechanical. A design turn produces a whole artifact: layout, styling and copy, all at once. A vague brief produces a wrong first version, so you ask for changes, and every "make the hero bigger" round costs about as much as the first draft did. Ten rounds later the page still has filler copy and the week is gone.

The setup that stops it is module 1 of the kit. The shape of it fits in five lines:

• Decide the exact deliverable before you open Design, down to the list of sections.
• Write the brand file once, so Claude never has to guess at colors, type or tone.
• Send one complete brief per artifact instead of building it up over many turns.
• Stop iterating inside Design after the second version. Small edits happen in the HTML or in Claude Code, where they cost far less.
• Export as HTML the moment a version is usable, so a bad later turn never costs you a good earlier one.

Whether Design usage and normal chat usage draw from the same weekly pool is the first thing to check on your own plan.

The DESIGN.md brand file

DESIGN.md is a single Markdown file that holds every fact about your brand a designer would ask for on day one. Name and one-line positioning. Who the audience is. The voice, in a few plain adjectives with an example sentence. The color palette as hex values, with what each color is for. The type pairing. A spacing and corner-radius rule. Logo usage. A short list of things never to do. Two or three reference sites you admire and why.

Without it, Claude falls back on its own defaults, and its defaults look like every other AI-generated page. With it, the first version arrives in your colors and your voice, and the ten rounds of "make it feel more like the brand" never happen. That is the return on 15 minutes of writing.

You attach or paste the file at the start of a design conversation. Module 2 walks through writing yours, and the vault includes five DESIGN.md starters plus a brand interview prompt that asks you the questions and drafts the file from your answers.

The brief that gets the first version right

Claude Design fills every gap in a brief with invention. Leave out the headline and it writes one. Leave out the pricing and it makes some up. Leave out the audience and it designs for nobody in particular. So the brief exists to close the gaps before Claude opens them.

At a high level, a brief that works has six parts:

1. The deliverable in one line: what it is, what it is for, where it will live.
2. The reader, and the single action they should take after seeing it.
3. The sections in order, with the real copy or the real numbers for each one.
4. A pointer to the brand file.
5. Constraints: viewport, dark or light, one page or several, no stock imagery, anything the output must never contain.
6. The export target, so Claude builds in a form that survives the trip out.

That is more writing up front than a one-line prompt, and it is exactly where the weekly limit gets saved. The vault holds 25 copy-paste briefs, one per deliverable type, so you edit rather than start from blank.

If you work in Claude Code, the same brief can become a reusable skill that runs the same way every time. That is what my Claude Skills mini course teaches, and the skills directory has free ones to start from.

Getting a design out of Claude Design

There is no native Figma export. Plan for HTML first and route everything else through that file.

Destination	Route	Watch for
Browser, or a developer handoff	Export the artifact as HTML	Keep every version. The HTML is the source of truth
Figma	HTML, then html.to.design	Layers arrive as the plugin interprets them, so tidy the frames after import
Production build	Hand the HTML to Claude Code with the handoff line from the vault	Tell it which parts are placeholder copy
Slides	PPTX with fonts intact	Check the fonts on the machine that will present
Print	PDF with bleed and crop marks	Put the bleed size in the brief, in millimetres

Module 8 covers each route step by step. The short version: export early, export as HTML, and treat every other format as a conversion of that file.