Claude Design 3.0 for GTM LinkedIn Assets

Sources: multiple documented Claude Design workflows for LinkedIn assets, including a Substack walkthrough on building a repeatable LinkedIn carousel skill (Annika Helendi, AI & Marketing), a guide on extracting a brand system from a website into Claude Design (AI Maker Substack), and hands-on carousel-building writeups from Friday AI Club and independent testers.

The consistent finding across every real-world writeup: Claude Design's actual sweet spot for GTM work is not one-off asset generation, it is building one on-brand template once, packaging it as a reusable skill, and then running that skill on any new topic to produce a finished asset in under a minute. The idea is rarely the hard part. The design and layout work is what kills most content before it gets posted, and that is the specific bottleneck this workflow removes.

4 modules:

1. Setting up the design system (the step everyone skips and pays for later)
2. Building the LinkedIn carousel template and exporting it correctly
3. Turning the template into a permanent, reusable skill
4. The honest limitations documented by people who have actually used it

Why the design system comes first, always

Skipping the design system step is the single most common reason Claude Design output looks generic. Without one, every asset defaults to Claude's own house style rather than the brand's. With one, colours, fonts, logo, and image style are locked in before a single template gets built.

The fastest way to build one: hand Claude Design your existing website URL and let it extract what is already there, then correct whatever it gets wrong, rather than describing your brand from scratch. If no website exists yet, a dedicated brand-extraction skill (covered in Module 1) can still produce a usable design system from whatever source material is available.

The step that determines whether every future carousel, deck, or post looks on-brand or generic. Two paths depending on whether a website already exists.

Path 1: extracting from an existing website

In the design systems panel, provide colours, fonts, logo, and image style directly, or hand Claude Design the website URL and let it read what is already live, then correct whatever comes back wrong. This is faster than describing a brand from memory and tends to catch details (exact colour values, spacing patterns) a manual description would miss.

Path 2: no website yet, using a brand-extraction skill

A documented approach for founders without an existing site: build a dedicated skill that runs inside Claude Cowork or Claude Code, point it at whatever source material does exist (a Notion page, a pitch deck, a set of brand notes), and have it generate a structured design-system output folder. That folder then gets uploaded into Claude Design during onboarding.

The genuinely useful detail in this approach: the skill is instructed to generate hard rules Claude treats as constraints, not suggestions. Not "the site uses dark colours" but "background is always #0D0D0D." Not "usually use Inter" but "always use Inter, never any other font." That level of specificity, rather than vague style guidance, is what makes output stay consistent across dozens of future generations rather than drifting.

# Skill: brand-system-extractor
mkdir -p .claude/skills/brand-system-extractor

Extract a design system from the source material provided.
Source: [WEBSITE URL, PITCH DECK, NOTION PAGE, OR BRAND NOTES]

Produce a structured output folder containing:
- Exact colour values (hex codes, not colour names or descriptions)
- Exact font names and where each is used (headings vs body vs captions)
- Logo files and usage rules (minimum size, clear space, backgrounds
  it can and cannot sit on)
- Image and photography style (tone, treatment, whether people are shown)
- A written list of hard constraints Claude Design must follow exactly,
  phrased as absolute rules, not stylistic suggestions
  (e.g. "Background is always #0D0D0D" not "the site uses dark colours")

Save as design-system/ ready to upload into Claude Design onboarding.

What the meta-skill unlocks

Once this extraction runs once, the resulting folder can also be installed as its own permanent Claude skill. From that point on, invoking a single named skill (your own brand's design skill) and describing what you need returns output that is already on-brand, without re-uploading the design system or re-explaining brand rules in every future session. One extraction, and every future session already knows who you are.

Once the design system exists, building the actual carousel template is a one-time job that pays off on every future post.

Building the template

Start a new design, select the design system already built, leave the template type field empty, attach a headshot if using one, and prompt for a carousel template rather than a finished carousel. The distinction matters: a template is a reusable structure, a finished carousel is a one-off asset that has to be rebuilt from scratch next time.

Ask for a few title-slide options alongside a mix of body layout types: a numbered point slide, a big-statistic slide, a screenshot slide, and a closing CTA slide. Having several layout variations available from the start is what prevents every future carousel looking identical regardless of topic.

"Using my design system, build a LinkedIn carousel template.
Give me 2-3 title slide options.
Body layout variations needed:
- A numbered point slide
- A big-statistic slide
- A screenshot slide
- A closing CTA slide
Attach my headshot to the title slide options.
This is a template for repeated future use, not a one-off finished carousel."

Exporting correctly

Export the template as HTML, not PDF. HTML is an editable structure that Claude Cowork or Claude Code can read, modify, and regenerate programmatically for future topics. A PDF is a flattened, final file with no editable structure behind it, which defeats the entire point of building a reusable template.

For a finished carousel (once the template exists and is being run against a real topic), export each card as an individual PNG rather than one combined PDF, and export the card content only, not the background. This produces files ready to upload directly to LinkedIn or Instagram as a native carousel post, one file per card.

A documented reference-image trick

Attaching a reference image of a carousel style you like, even loosely, meaningfully changes the output quality compared to a text-only description. This is optional but consistently reported as making a real difference, since visual style is easier to show than to describe precisely in words.

The step that turns a one-time template build into a system you never have to rebuild: packaging the finished HTML template as a permanent Claude skill.

The packaging workflow

Once the template is exported as HTML, bring that file into Claude Cowork or Claude Code and have it extracted into a reusable skill. From that point forward, invoking the skill with a new topic produces a finished, on-brand carousel using the exact layout variety already built into the template, without touching Claude Design again for that asset type.

# Skill: linkedin-carousel-builder
mkdir -p .claude/skills/linkedin-carousel-builder

Use the exported HTML carousel template at [PATH TO TEMPLATE].
Topic: [YOUR TOPIC]

Produce:
1. A finished carousel using the template's title slide, body layout
   variations (numbered point, big-statistic, screenshot, closing CTA),
   and brand design system exactly as built.
2. 5 LinkedIn post caption options, each written from a different angle
   for the same carousel (a question hook, a contrarian take, a specific
   number, a story opener, a direct audience callout).
3. A first-comment CTA to pair with the post.

Export each carousel card as an individual PNG, card content only,
no background, ready to upload directly to LinkedIn.

Apply brand-voice.md if available. No em dashes. No hashtags in captions.

Why this is the actual unlock for GTM teams

The two expensive steps (building the design system, building the template) each happen once. The third step, running the skill on a new topic, is the one that repeats forever, and it is the only one that takes under a minute once the first two are done. This is what separates a repeatable content system from a one-off demo: the cost of producing carousel number 50 is the same as the cost of producing carousel number 2.

Optional: pushing to Canva instead of publishing directly

For teams that want a human polish pass before publishing, Claude can connect directly to a Canva account (Claude Pro required for connectors; Canva Pro required for Claude to edit existing files rather than only create new ones) and push a drafted carousel or design concept into Canva for final editing, rather than exporting straight to PNG for immediate posting. Useful when the workflow includes a designer or brand reviewer before anything goes live.

What people who have actually used this for real GTM work report, stated plainly rather than glossed over.

Token cost and iteration

Getting to a genuinely decent first carousel set has been reported to take at least a few iterations, which burns through tokens faster than a single-shot generation would suggest. The template-once, skill-forever workflow in Module 3 exists specifically to absorb this cost at the template-building stage rather than paying it again on every single future carousel.

No native image generation

Claude Design does not generate original images from a prompt. Screenshots, photos, and existing graphics have to be supplied and dropped in manually; Claude Design lays them into the design but does not create new photographic or illustrated imagery on its own. Plan for this by having source screenshots and images ready before starting a design session rather than expecting the tool to generate them.

Text-heavy content is the strength, not visual complexity

Claude Design is consistently reported as strongest for text-based carousels (a numbered list, a framework, a set of takeaways) and comparatively weaker for carousels that lean heavily on custom images or screenshots, since dropping in and precisely positioning external images is more fiddly than generating clean text layouts from the template.

When to still use a template tool like Canva instead

For teams running very high carousel volume where token cost and iteration time genuinely matter more than the on-brand automation benefit, a traditional template tool with a fixed layout can still be faster for pure volume production. The honest framing from people who have tested both: Claude Design is a strong starting point for inspiration and for the template-to-skill workflow specifically, but output should still go through a human review pass before publishing, the same as any AI-generated creative asset.

The full workflow and both skills, condensed.

The full sequence

# STEP 1: Design system (once)
# In Claude Design's design systems panel:
# Either hand it your website URL and correct what comes back wrong,
# or run the brand-system-extractor skill from source material
# and upload the resulting folder during onboarding.

# STEP 2: Carousel template (once)
"Using my design system, build a LinkedIn carousel template.
Give me 2-3 title slide options.
Body layouts: numbered point, big-statistic, screenshot, closing CTA.
Attach my headshot to the title slide options."
# Export as HTML, not PDF.

# STEP 3: Package as a skill (once)
# Bring the HTML template into Claude Cowork or Claude Code.
# Extract it into a permanent skill: linkedin-carousel-builder

# STEP 4: Run forever (every future topic)
"Use my LinkedIn Carousel Builder skill. Topic: [YOUR TOPIC]."
# Returns: finished carousel PNGs, 5 caption options, first-comment CTA.

Both full skills in one place

# SKILL 1: brand-system-extractor (run once, no website needed)
mkdir -p .claude/skills/brand-system-extractor

Extract a design system from: [WEBSITE URL / PITCH DECK / BRAND NOTES]
Produce: exact hex colour values, exact font names by use case, logo
usage rules, image/photography style, and a list of hard constraints
phrased as absolute rules (e.g. "Background is always #0D0D0D").
Save as design-system/ ready for Claude Design onboarding.

# SKILL 2: linkedin-carousel-builder (run on every future topic)
mkdir -p .claude/skills/linkedin-carousel-builder

Use the exported HTML template at [PATH].
Topic: [YOUR TOPIC]
Produce: finished carousel (title + numbered point + big-stat +
screenshot + closing CTA layouts), 5 caption angle options, one
first-comment CTA. Export each card as an individual PNG, card only,
no background. No em dashes, no hashtags in captions.

What to expect realistically

Budget several iterations and real token spend for Steps 1 and 2. Step 4 is genuinely fast, under a minute per new carousel, once the first two steps are done properly. Plan to supply your own screenshots and images rather than expecting generation. Run a human review pass before publishing, same as any AI-assisted creative output.