Session 4 · Creative studio: branding, marketing, design

Part 0

Why this session exists

Marketing's bottleneck was never ideas - it's production and consistency. Every caption re-litigates the brand voice, every landing page waits for an engineer, every design review is opinion versus opinion. Today we install systems for all three: a brand voice that travels, a content engine that multiplies, and a critique loop with actual standards.

Live - presented in session Self-study - read after class ★ Build-along demo Applies Claude 101 + skills from Session 3

★ What you walk out with today A Brand HQ project with YOUR extracted voice guide, a working landing-page prototype you built and iterated live, and a case-study interview kit that turns project wins into sales assets.

Part 1 · brand as a system

Lock the voice once, use it everywhere 5 min live

"Sounds like us" should be infrastructure, not a review comment. The move: extract the voice from your BEST past work, write it down as rules, install it as a project (Session 2) or skill (Session 3).

LiveExtract your voice from evidence, not adjectives3 min▶

Most brand-voice docs say "confident, human, bold" - which describes every brand on Earth. The fix is archaeological: feed Claude your 5 best-performing pieces and make it reverse-engineer the ACTUAL patterns.

★ The voice extractorHere are 5 pieces of our best content - the ones that felt most "us" and performed best. [paste them] Reverse-engineer our brand voice as a working guide: 1. Sentence mechanics: typical length, rhythm, how we open, how we close. 2. Vocabulary: 10 words/phrases we clearly love; 10 we never use (infer the banned list). 3. Tone moves: how we handle claims (bold? hedged?), humor (type and dose), the reader ("you"? "we"? distance?). 4. Three before/after examples: a generic sentence rewritten in our voice. 5. The one-paragraph "voice in a nutshell" a freelancer could follow. Be specific enough that two different writers using this guide would produce interchangeable drafts. Flag any inconsistencies you noticed between the 5 pieces.

Real world

A B2B team ran this on their five best LinkedIn posts and discovered their real voice was "short declaratives, one number per post, no exclamation marks, gentle self-deprecation in the last line" - none of which was in their official brand book. The extracted guide went into Brand HQ; the next month of posts was written 3x faster and no one could tell which were AI-first drafts.

Install the output in your Brand HQ project (from Session 2) as knowledge + instructions - or as a brand-voice skill (Session 3) so it fires in ANY chat. Voice + banned words + best examples: that trio is the system.

🔍 Click to zoom - one project needs project knowledge; every chat needs a skill

Self-studyWhat Claude can and can't do for visuals - the honest map2 min read▶

🔍 Click to zoom - Claude drafts and critiques; the camera and the judgment call stay human

Claude does brilliantly	Claude does NOT do
Copy, concepts, scripts, briefs, naming	Generate photographic images (pair with your approved image tool)
Critique designs from screenshots (it sees images)	Edit image files pixel-by-pixel
Working web pages, interactive mockups, diagrams as artifacts	Native .fig files - it prototypes, your designer refines
Decks and docs via the pptx/docx skills	Replace brand judgment - Discernment stays yours

The workflow that wins: Claude writes the image BRIEF and the prompt for your image tool, critiques the results against brand, and assembles everything into the deliverable. It's the creative director's assistant, not the camera.

Part 2 · the content engine

One asset in, a week of content out 4 min live

The atomization pattern: produce one substantial "core asset", then let Claude cut it into every format your channels need - each in the right native shape, all in the locked voice.

🔍 Click to zoom - the atomization engine: one core asset, five native formats

LiveThe atomization prompt + campaign variants2 min▶

★ The atomizer (run inside Brand HQ)Here is our new core asset: [paste case study / report / webinar transcript] Cut it into 5 channel-native pieces, each in our brand voice (use the voice guide in project knowledge): 1. LinkedIn post (150 words max, our hook style) + a 6-frame carousel outline 2. A 3-email nurture sequence (subject lines + 80-word bodies, one idea each) 3. A sales one-pager: problem, approach, results with numbers, one client quote 4. A 60-second video script with caption text 5. Five exec talking points for leadership to use verbatim Rules: each piece must stand alone; re-argue for the format, don't summarize; keep every number exactly as sourced; flag any claim that needs verification before publishing.

Variants for testing: "Give me 5 hooks for piece 1: one curiosity, one contrarian, one number-led, one story, one question. Rank them for OUR audience and say why." You choose with judgment; it supplies the spread. (A/B discipline, minus the blank page.)

Self-studyCase studies that sell - the interview kit2 min read▶

The strongest core asset is a case study, and the hard part is extraction, not writing. Run this interview on the project owner (or paste the project docs) - it covers the official "write case studies" playbook:

★ Case-study interview kitI want to write a client case study. Interview me one question at a time (max 8 questions) to extract: - The client's situation and the stakes (what happened if nothing changed?) - The trigger moment they decided to act - What we actually did, in plain language, step by step - Numbers: before/after metrics, timeline, effort saved (push me for specifics) - One direct quote-worthy sentiment from the client - What surprised us - the detail that makes it credible Then draft the case study: headline with the result in it, 400-600 words, structure: situation → approach → results (numbers first) → client quote → one-line takeaway. Flag every number and claim I need to verify with the client before publishing.

Part 3 · the design improvement loop

Critique with standards, not vibes 4 min live

Claude reads images. Screenshot anything - a slide, a banner, a landing page - and run a structured critique against named principles instead of "make it pop".

LiveThe structured critique prompt2 min▶

★ The design critique (attach a screenshot)Critique this design as a senior art director. Structure: 1. FIRST IMPRESSION - what does a viewer get in 3 seconds? Is that what we want? 2. HIERARCHY - what's fighting for attention? What's the ONE thing, and does everything else support it? 3. Score against: contrast, alignment, whitespace, type consistency, color discipline (one accent doing one job), accessibility (text sizes, contrast ratios). 4. BRAND FIT - check against our voice/visual rules [in project knowledge]. 5. The 3 highest-impact fixes, ranked by effort-to-impact - specific enough that I could brief a designer in one line each. Do NOT redesign it wholesale. Do NOT soften: if the hierarchy is broken, say so.

Then the loop from Session 1 applies: fix, re-screenshot, re-critique. Two laps is usually the difference between "fine" and "ship it".

Real world

A team ran their event banner through the critique: "three type sizes competing, CTA has the least contrast on the page, logo doing nothing at that size." Fifteen minutes in their design tool later, signups from the banner doubled. The critique wasn't genius - it was consistency: the same named standards, applied every time, with no meeting.

Part 4 · website prototypes without engineers

From brief to clickable page in one session 3 min live

Artifacts (Session 1) turn Claude's output into working web pages beside the chat - real HTML, real interactivity, sharable by link. For marketing, that means the landing-page conversation changes from "wait for eng" to "react to a working draft".

LiveWhat prototypes are FOR (and not for)3 min▶

For: testing a message and layout on a real-feeling page, stakeholder sign-off ("this, but the hero says X"), campaign microsites for review, interactive mockups (calculators, quizzes) to spec the real build.
The iteration loop is the value: "hero too corporate - warmer", "move proof above the fold", "add a 3-tier pricing table" - each takes seconds, live, in front of stakeholders. Ten iterations in a meeting beats two per sprint.
Handoff: when approved, the artifact IS the spec - engineers get working HTML plus your iteration history instead of a wireframe PDF. (Publishing it properly - hosting, domain, analytics - is still an engineering/IT step; Session 5 crowd can ship it to GitHub Pages in minutes.)
Not for: production sites with forms collecting real customer data - prototypes prove the design, production needs the proper pipeline.

Demo 1 of 3

Build Brand HQ + extract your voice ★ 8 min · everyone builds

Create (or open) your Brand HQ project. Gather your 5 best pieces of content - posts, emails, pages that felt most "us".

Run the voice extractor (Part 1) on them. Read what comes back - expect at least one pattern you never noticed you had.

Correct one thing it got wrong (Discernment lap), then save the final guide into the project's knowledge + instructions: "All drafting in this project follows the voice guide."

Prove it: "Draft a LinkedIn post announcing [something real] in our voice." Compare against your gut. Tighten the guide if it missed.

Demo 2 of 3

Landing page, live, iterated by the room ★ 9 min · follow along + shout edits

In Brand HQ, run the brief: "Create a landing page artifact for [a real upcoming campaign]: hero with our value prop, 3 proof points, one testimonial block, pricing teaser, single CTA. Our voice. Modern, clean, mobile-friendly."

The room drives iteration - collect three reactions and apply them live: "warmer hero", "CTA above the fold", "add the logos strip". Watch each land in seconds.

Interactive upgrade: "Add an ROI mini-calculator: [inputs] → [simple formula] → result line." Now it's a page a static mockup could never be.

Share the artifact link with the room. That link is the stakeholder review, the spec, and the handoff - one object.

Data tip Use public or placeholder numbers in prototypes - a shared artifact travels further than you plan. Real pricing goes in at production time, through the proper pipeline.

Demo 3 of 3

Critique a real asset from this quarter ★ 5 min · everyone runs one

Everyone screenshots one real asset they shipped recently - a slide, a banner, an email header.

Run the structured critique (Part 3) on it. Two quiet minutes.

Round-robin: each person shares their single highest-impact fix. Notice how consistent the standards are across totally different assets - that consistency is what you just installed.

After the session

This week ◐ 45 min total

Finish Brand HQ: add the banned-words list and your 3 gold-standard examples. Share the project with the team - one voice, everyone.
Run the atomizer on your best existing case study or report. Publish at least two of the five cuts.
Turn the critique into a skill (Session 3 method): design-critique with YOUR brand rules baked in - now the standard applies itself.
Interview one project owner with the case-study kit. Verified draft to the client for approval by Friday.
Iterate the landing page to done, then hand the artifact link to engineering - ask them how much spec-writing you just skipped.

Source material

Official sources covered

Session 4 is the application session: it operationalizes Claude 101's Projects + Artifacts for creative work (taught in Session 1, applied here), plus the official "write case studies" use-case guide.

✓Claude 101 - Projects & Artifacts, appliedBrand HQ system + landing-page prototyping workflow

✓"Write case studies" use-case guidethe interview kit + structure in Part 2

◐marketing-skills community packinstalled via Session 3's shelf - copywriting/ads/email skills slot into this workflow

Check yourself

Three questions before you go 🎯 ◐ 90 seconds

1 · The right source for your brand voice guide is...

Evidence, not adjectives. The extractor finds the patterns you actually have - sentence mechanics, banned words, tone moves.

2 · Atomizing a case study into a LinkedIn carousel means...

Each format stands alone and is re-argued for its channel. A carousel is not a chopped-up blog post.

3 · What can Claude honestly do for your visuals?

It sees images and reasons about them brilliantly; it doesn't generate photos. Pair it with your approved image tool.

Session 4 cheat sheet · pin this

Voice systemExtract from 5 best pieces → rules + banned words + examples → Brand HQ project. Evidence, not adjectives.

AtomizeOne core asset → 5 native formats. Re-argue per format, never summarize. Numbers stay exact.

Critique order3-second impression → hierarchy → contrast/alignment/whitespace/type/color → brand fit → 3 ranked fixes.

Prototype loopBrief → artifact → room reacts → iterate in seconds → share link = spec + sign-off.

Case study spineSituation → approach → results (numbers first) → quote → takeaway. Interview, then draft.

Honest limitsClaude writes briefs and critiques images; it doesn't generate photos. Pair with your image tool.