Jonathon's AI Wiki

Module 5 — Claude Design

18 min read

Source: Anthropic tutorial: Claude Design for presentations + Anthropic tutorial: Claude Design for prototypes and UX + Paul Couvert walkthrough (YouTube, 14 min)

Time: Read 3 min | Watch 20 min | Practice 15 min

Claude has a separate product for visual work: Claude Design, at claude.ai/design. Launched April 2026 as the first Anthropic Labs product. Built for making designs, decks, landing pages, prototypes, and one-pagers from text prompts.

Watch First

Three sources. Do them in order.

1. Paul Couvert — “Claude Design: Learn 80% in a Few Minutes” (~14 min, YouTube)

Direct: youtube.com/watch?v=GPqcZ_GSqOg

The single best first-time walkthrough of Claude Design. Covers the 4 creation tabs, the 5-minute design system setup, the “Decide for me” Q&A flow, and the editing trio (Tweaks / Edit / Comment). Start here.

This is also the source that closed the question on Claude Design usage limits: it has a separate weekly pool from main Claude, heavy first-time users hit the cap in ~2 hours of exploration, and motion graphics burn through the pool faster than static designs. Plan accordingly.

2. Anthropic tutorial: Design for presentations and slide decks (~5 min read)

Direct: claude.com/resources/tutorials/using-claude-design-for-presentations-and-slide-decks

Anthropic’s own guide to producing slide decks — the most WEO-relevant use case (client pitches, QBRs, one-pagers).

3. Anthropic tutorial: Design for prototypes and UX (~5 min read)

Direct: claude.com/resources/tutorials/using-claude-design-for-prototypes-and-ux

For web design and landing page mockup work.

4. Peter Yang — “Claude Design: Everything You Can Build in 16 Minutes (5 Real Use Cases)” (YouTube, 333k views)

Direct: youtube.com/watch?v=WMnk1LFBMqA

Same length as Paul Couvert’s primer (~16 min) but organized around five specific deliverables instead of the feature tour. Watch this OR Paul Couvert; the two together is ~30 min and overlaps. Peter’s picks lean toward client-facing artifacts (decks, landing pages, one-pagers) which match WEO’s daily output closely. Full channel profile in Module 8.

5. Nate Herk — “Claude Design Builds Beautiful 3D Websites Instantly (full tutorial)” (YouTube, advanced)

Direct: youtube.com/watch?v=TcFeSjwTo7g

Hands-on build of a high-end 3D marketing site end-to-end in Claude Design. Watch after Paul Couvert if you want to see what the ceiling looks like — the kind of output that earns premium landing-page budgets. Not required, but a useful “what’s possible” reference before you pitch Claude Design work to a client.

6. RoboNuggets — “Claude Code + Karpathy Slides Just Changed Presentations Forever (MARP System)” (YouTube, 64k views — WEO-relevant)

Direct: youtube.com/watch?v=RBcc_ezfh1s

This one is especially WEO-relevant: the course you’re reading ships its 9 slide decks via exactly this Marp-first pipeline (markdown source → npx @marp-team/marp-cli → PDF → Cloudflare static asset). If you want to see how the sausage is made, this video walks the pattern end-to-end. Useful for anyone building repeatable deck templates or documenting a client playbook that needs to render as both web page and PDF. Full RoboNuggets channel profile in Module 8.

7. RoboNuggets — “Claude JUST Changed Design as We Know It (10+ Use Cases)”

Direct: youtube.com/watch?v=GaQIamkXJeo

Design-use-case menu in the same format Peter Yang’s 16-min tutorial uses, but RoboNuggets leans heavier into cross-tool combinations (Claude + Excalidraw, Claude + NotebookLM, Claude + image gen). Skim for ideas to try after you’ve done Modules 5’s core tutorials.

Why It Matters at WEO

The combination of “produces client-facing decks” + “produces landing page mockups” + “integrates into existing Claude account” means Claude Design can replace or substantially compress the Canva → Figma → Google Slides workflow for a meaningful slice of WEO’s client-facing output. Pitch decks, landing page mockups, social graphic sets, infographic summaries, one-page proposal summaries. Faster than Canva on a good day, better than a junior designer on a bad day.

The limit: you’re not replacing professional designers on brand-sensitive or high-volume work. You’re replacing the 30% of design work that’s routine-but-still-takes-time.

WEO-Specific Overlay — When to Use Claude Design vs. Regular Claude.ai

Use Claude Design when the deliverable is…Stay on regular Claude.ai when the deliverable is…
A pitch deck or QBR presentationA blog post, email, or long-form doc
A landing page mockup (pre-engineering)Research, strategy, planning notes
A social post graphic setAd copy, meta descriptions, outlines
A one-page proposal summary with visual hierarchyInternal memos, briefs, content calendars
A competitor-teardown visualA competitor content teardown (text analysis)

Rule of thumb: if the deliverable has a visual hierarchy and you’d otherwise open Canva, Figma, or PowerPoint for it — try claude.ai/design first.

WEO-Specific Overlay — Five Dental Design Use Cases

Lifted from the broader community playbook, filtered for WEO work. These are copy-paste prompts — fill in the bracketed placeholders with client details:

  1. Client pitch deck (QBR).

    “Create a 10-slide pitch deck for [CLIENT PRACTICE] for their quarterly business review. Include sections for wins this quarter, KPIs (placeholder numbers), upcoming campaigns, and strategic roadmap. Practice is a [GENERAL / PEDIATRIC / COSMETIC / FAMILY] dental practice in [CITY]. Brand voice: [WARM / MODERN / PREMIUM].”

  2. Landing page mockup.

    “Design a landing page for a dental practice’s Invisalign service. Hero with headline + subhead + CTA, 3 benefit callouts with icons, one-paragraph ‘how it works’ section, testimonial section with 2 patient quotes (placeholder), financing mention, primary booking CTA. Warm consumer-friendly tone — NOT clinical.”

  3. Social graphic set.

    “Design 5 Instagram posts for a family dental practice — one per service (cleanings, whitening, implants, pediatric, emergency). Consistent visual system. Each post has one benefit-led headline (max 8 words) + one supporting line + practice logo placeholder + CTA ‘Book today.‘”

  4. One-page proposal summary.

    “Turn this 12-page proposal into a 1-page executive summary with key numbers, scope, and timeline. Visual, not text-heavy. Use a header, 3 benefit pillars, a timeline strip, a numbers bar, and a CTA. Tone: confident, grounded.” (Upload the PDF first in the same chat.)

  5. Competitor capture → our version.

    “Look at this competitor landing page screenshot. Design a stronger version for [CLIENT] that covers the same benefits but with warmer brand voice and clearer pricing communication. Use our brand colors [HEX CODES].” (Upload the screenshot first.)

WEO-Specific Overlay — Working Methods

Three patterns that make Claude Design compound for the team (all from Paul Couvert’s walkthrough and Anthropic’s tutorials):

  1. Set up a design system once per major client. Colors, fonts, logo, voice tokens. 5-minute setup. Everything you produce in that project inherits it. Stops you re-specifying brand every time. Deep dive below.
  2. Use templates as primitives. When you ship a deck you’re proud of, save it as a template. Next time, start from the template instead of from scratch.
  3. “Decide for me” in early iterations. When Claude asks clarifying questions and you don’t have a strong opinion, say “decide for me.” Faster than trying to answer every question up front — you can always change it after seeing the first draft.

Setting Up a Design System — Dental Deep Dive

Claude Design’s 5-minute onboarding asks for brand assets once; everything you produce afterward in that design context inherits them. Most WEO team members rush this and regret it. Slow down for 10 minutes on setup; save hours across every downstream deliverable.

The six fields that actually matter for dental work

Claude Design’s onboarding covers more than this, but these are the six where getting it right compounds:

  1. Primary colors (hex codes, named). Not “the practice’s blue” — the actual hex code, and what it means. Example for Smile Springs: #2B5F87 "Primary — trust blue", #F7941D "Accent — warm action (CTAs only)", #F4F1EA "Background — paper". Claude Design respects named colors across decks.
  2. Typography pair. One display, one body. Not a font system; the two fonts. For Smile Springs: “Display: Fraunces (editorial serif, warm). Body: Inter (clean sans, readable).” If the client has a brand font they actually own, use it; if they don’t, pick two Google Fonts and stick with them across all deliverables.
  3. Logo + safe-zone rules. Upload the logo with transparent background. Note minimum size. Note banned backgrounds (“never on #F7941D — contrast fails WCAG”).
  4. Voice tokens — short, mid, long. Three example phrases that sound like this brand. For Smile Springs: Short — “Saturday smiles.” Mid — “A family-first dentist in Columbus, booking in under a minute.” Long — “We’re the dentist your grandma calls ‘that nice young practice on High Street’ — not the one who tries to upsell whitening mid-cleaning.” Voice tokens teach faster than voice instructions.
  5. Imagery rules. For dental: “Real patient photography always. Never stock. Never AI-generated people. Smiling faces OK; clinical close-ups avoided. Families and non-staged candid over posed.” If the client has a photo library, connect it to the design project.
  6. The “never” list. Brand-specific bans. For Smile Springs: “Never use: gradients that cross blue-to-orange, cursive fonts in CTAs, before/after clinical close-ups in marketing contexts, emoji in headlines.”

Smile Springs Family Dental — worked design-system example

design_system: "Smile Springs Family Dental"
 
colors:
  primary:
    hex: "#2B5F87"
    name: "Trust blue"
    use: "Headers, primary UI, backgrounds for hero sections"
  accent:
    hex: "#F7941D"
    name: "Warm action"
    use: "CTAs only. Never backgrounds, never body text."
  support:
    hex: "#F4F1EA"
    name: "Paper"
    use: "Page backgrounds, card fills, space at rest"
  ink:
    hex: "#1C2B3A"
    name: "Ink"
    use: "Body text on light backgrounds"
 
typography:
  display: "Fraunces (warm editorial serif — H1, H2)"
  body: "Inter (clean sans — paragraphs, UI)"
  forbidden: "No Comic Sans (obvious), no Times New Roman (feels dated), no Montserrat (overused in dental)"
 
voice:
  short: "Saturday smiles."
  mid: "A family-first dentist in Columbus, booking in under a minute."
  long: >
    We're the dentist your grandma calls "that nice young practice on
    High Street" — not the one who tries to upsell whitening
    mid-cleaning.
 
imagery:
  default: "Real patient photography (client-provided)"
  fallback: "Warm candid stock — never overtly staged"
  avoid:
    - "AI-generated people"
    - "Before/after clinical close-ups in marketing context"
    - "Beach / vacation imagery (no thematic relevance to dental)"
 
never:
  - "Gradients that cross blue-to-orange"
  - "Cursive fonts in CTAs"
  - "Emoji in headlines"
  - "All-caps outside of kicker labels"
  - "Drop shadows on text"

Load that once into the Smile Springs design project. Every deck, landing page, and social asset you produce in that project inherits it. Claude stops asking “what colors do you want?” because it already knows.

Pre-Client Deck Review Checklist

Before sharing any Claude-generated deck with a client, run this 10-point review. Catches 90% of what would otherwise be caught the embarrassing way — by the client.

  1. Claim check. Every statistic, every clinical claim, every client-specific data point: is it real, sourced, or flagged as placeholder? Claude Design will confidently invent plausible numbers. Verify or mark “DATA TBD”.
  2. Brand-guide compliance. Banned phrases absent? Colors on-palette? Typography on-system? Imagery guidelines followed? (If you set up the design system, this should be clean — but verify.)
  3. Patient-privacy check. No real patient names, no recognizable patient photos, no actual testimonials you haven’t cleared, no before/after imagery that includes identifiable faces. Module 7 governs — this is the do-not-ship risk.
  4. Tone calibration. Read the deck out loud. Does it sound like the client? Or does it sound like a design tool’s default voice? If it reads as Claude’s default — fix Role + brand voice in the prompt and regenerate.
  5. Internal consistency. Do headlines, subheads, and body copy point in the same direction? Claude Design can produce slides where each is individually strong but they disagree about the story.
  6. Visual hierarchy. Every slide: can you identify the primary thing the viewer should look at first? If every slide has the same emphasis pattern, the deck reads flat — vary visual weight intentionally.
  7. Pagination and flow. Do transitions make sense? Does the deck have an arc (setup → reveal → next steps) or is it a list of slides in arbitrary order?
  8. Export test. Export to PDF. Open in Preview / Acrobat. Any text cut off? Any color shifts between the in-app preview and PDF? Any fonts that didn’t embed?
  9. Accessibility quick-check. Contrast between text and background passes WCAG AA at minimum. Interactive elements (if it’s an interactive HTML deck) have visible focus states. No color-only-conveyed information.
  10. Does it answer the client’s question? The deck was made to solve a specific problem (QBR, pitch, proposal). Does it actually solve that problem? Or did it drift into looking-good-without-answering?

Common Failure Modes and Recovery Moves

Claude Design’s most-missed failure modes — and how to recover without starting over:

Failure 1 — The deck feels “AI-designed”

Obvious over-polished sameness. Every slide has 3 columns or 3 bullets. The deck looks like every other Claude Design deck.

Recovery: Tell Claude to vary. “Break the 3-column pattern. Use full-bleed imagery on slides 3 and 7. Use a single-number hero slide for the KPI page. Make the closing slide almost empty — just one line and a CTA.” Variety has to be requested; default Claude produces template rhythm.

Failure 2 — Generic stock-style imagery

Every image looks like a generic stock photo. No specificity to the brand.

Recovery: If you have real client photos, upload them and reference them in the prompt. If you don’t, get more specific in image prompts: “Image for slide 4: an empty dental chair in a warmly-lit small practice waiting room. Morning light, plants nearby, not clinical. Warm wood tones.” Specific beats generic every time.

Failure 3 — The headline is right but the subhead repeats it

Claude writes “Why Smile Springs” as the headline and “Why Smile Springs is different” as the subhead.

Recovery: Tell Claude the rule: “Subheads must add information, not restate the headline. If the headline makes a claim, the subhead delivers evidence or a consequence.” Or use the edit icon on specific slides and rewrite the subhead yourself — faster than re-prompting.

Failure 4 — Too much content per slide

Claude treats slides like pages. 5 bullets, a sub-bullet per bullet, a footnote.

Recovery: Rule of thumb per slide: “One claim. Up to three supporting elements. Everything else becomes a new slide.” Tell Claude: “Reduce to one headline + up to three supporting items per slide. Split overloaded slides into multiple slides rather than cramming.”

Failure 5 — Wrong deck length

You asked for 10 slides; Claude produced 18. Or you asked for 10 and got 6 because Claude “consolidated.”

Recovery: Slide count constraints require reinforcement. “Deck must be exactly 10 slides — not 9, not 11. If content requires more, cut content. If content is thin, expand with a closing-next-steps slide.” Claude tends to drift toward its default deck length unless you pin the count.

Failure 6 — Motion graphics that don’t render right in export

Animations look great in the interactive HTML view but export to PDF as frozen odd-angle frames.

Recovery: Match deliverable to medium. Static deliverables (client email attachment, print) → prompt for static design. Interactive deliverables (screen-shared live presentation) → interactive HTML is fine. For client-shareable artifacts default to PDF-first design, not animation-first. Motion is nice-to-have, not load-bearing.

Iteration Patterns Beyond Tweak / Edit / Comment

The three built-in edit modes handle 80% of iteration. For the other 20%:

The “regenerate this slide only” pattern

Long deck, one slide is wrong. Don’t regenerate the whole deck. Open that slide’s context in the chat, request a regeneration with specific changes, leave everything else untouched.

The “show me three versions” move

When you’re not sure what direction is right, ask for three variants of a specific slide: “Give me three alternative versions of the hero slide — one emotional, one factual, one contrarian. Same info, different framing.” Pick the one that works. Saves you from revising a single version through three rounds.

The “steal this layout” move

If you see a reference design you like (a competitor’s deck, a well-designed report), paste a screenshot and say: “Use this layout structure for slide 4 of our deck, but with our content and brand.” Visual reference transfers faster than layout-describing.

The “rewrite the narrative, keep the design” move

After the design is locked but the copy needs revision: “Rewrite all headlines and body copy across all slides. Keep the design system, layouts, and image placements exactly as they are. Only text changes.” Lets you iterate on messaging without disturbing design work you’ve already approved.

The “handoff to Claude Code” move

When the landing-page mockup is locked and needs to become real code: “Package this design as a handoff bundle for Claude Code. Include the design tokens, component markup, and a spec for a React + Tailwind implementation.” Claude Design produces a bundle, you open Claude Code, point it at the bundle, and the real implementation gets built with the design system intact.

When Claude Design Is the Wrong Tool

Not every visual deliverable should go through Claude Design. Three cases to stay away:

  • High-volume production work under a print deadline. If you’re producing 40 ad variants for a campaign launch tomorrow, Canva (with the client’s brand kit loaded) is faster and more predictable. Claude Design is for shape-giving; Canva is for batching.
  • Brand-system work. Don’t use Claude Design to create a brand system from scratch for a new client. That’s design strategist / senior designer work. Claude Design is for applying an existing brand system to deliverables — not inventing one.
  • Anything a professional designer is already working on. If WEO has a designer in the loop for this client, running Claude Design in parallel creates coordination pain and usually produces work your designer has to redo. Defer to the designer; use Claude Design for the work the designer can’t prioritize.

Template Library Starter Set

As you ship work you’re proud of, save it as a template. These five are worth building first:

  1. Client QBR template. 10 slides: title, exec summary, wins, KPIs with placeholder numbers, losses/learnings, next-sprint plan, asks, appendix. Once built, every client’s QBR starts from this shell.
  2. Pitch deck template (new business). 12 slides: problem, competitive landscape, our approach, case study, team, process, pricing, next steps. Fill-in for each prospect.
  3. Landing page mockup — service page template. Hero / 3-benefit row / how-it-works / testimonial / financing / CTA. Use for every service landing page.
  4. Social graphic set template. 5-post Instagram carousel. One per service or campaign. Consistent visual system built in.
  5. One-page proposal summary template. For boiling a 12-page proposal down to an executive-friendly single page. Header, 3 benefit pillars, timeline strip, numbers bar, CTA.

Saving a template takes 2 minutes per deck after you finish it. Worth it every time.

Known Limits (don’t learn these on a deadline)

  • Separate weekly usage pool. Not shared with your main Claude.ai usage. First-time heavy users hit the cap in ~2 hours. Motion graphics burn faster than static.
  • Enterprise plans: Claude Design is off by default and your admin has to enable it. Check with WEO IT if claude.ai/design returns a permissions error.
  • Motion graphics are newer. Capable but less predictable than static work. Budget more iteration on anything with animation.

Key Takeaways

  • Claude Design = claude.ai/design. Separate product, same account, same login. Built for visual output.
  • Paul Couvert’s 14-min video is the single best first exposure. Watch it.
  • Use Claude Design when the deliverable is visual. Stay on regular Claude.ai for text-first work.
  • Invest 10 minutes in design-system setup per client — the six fields that matter for dental are colors, typography, logo, voice tokens, imagery rules, and the “never” list.
  • Pre-client review checklist is 10 items — run it before sharing. Claim-check and patient-privacy are the do-not-ship ones.
  • Six common failure modes have known recovery moves. Learn to diagnose from the output shape.
  • Iteration beyond Tweak/Edit/Comment: “three versions,” “steal this layout,” “keep design, rewrite copy,” Claude Code handoff.
  • Not every visual deliverable goes through Claude Design — high-volume production, net-new brand-system work, and parallel-with-designer work stay out.
  • Save templates as you ship. Five starters: QBR, pitch, service-page LP, social carousel, one-page proposal summary.
  • Watch the separate weekly limit, especially with motion graphics.

Try It (15 min, hands-on)

  1. Watch Paul Couvert’s video (14 min) — youtube.com/watch?v=GPqcZ_GSqOg.
  2. Open claude.ai/design. If you hit a permissions error, ping WEO IT — Enterprise plans disable Claude Design by default.
  3. Pick “Presentations.”
  4. Prompt:

    “Create a 5-slide internal presentation titled ‘Why Our Team Is Adopting AI.’ Audience: WEO marketing staff. Cover: why now, what changes, ground rules, first steps, resources. Tone: confident, not salesy.”

  5. Use “decide for me” on clarifying questions you don’t have strong opinions about.
  6. Once the deck is ready, try one of each edit mode: one Tweak (e.g., “make the palette warmer”), one Edit (directly change one headline), one Comment (annotate a slide).
  7. Export to PDF. You now have a reusable artifact for future team-onboarding conversations.

Done? Move on to Module 6.

Graph View

Backlinks