DAY 1 · BRAND SAFETY & VOICE
‹ Day page
→ next  ← back  F full
CODED · AI in Marketing

Brand Safety
& Voice

Feel the grind, red-team the unsafe output, and calibrate a brand voice your AI can't break — all inside Claude.

AI in Marketing Content Creation Day 1 of 3 · with Hassan
Press → to begin
The team today

This is your content team today.

Brand Leadowns the voiceCopywriterwrites the wordsDesignermakes the visualsVideo Editorcuts the clipsSocial Leadfits each platformPlannerschedules the week
A week of content takes a whole team — and every handoff is a place it can drift off-brand.
The move

Move every seat into Claude.

Brand Leadowns the voiceVoiceskillCopywriterwrites the wordsWritingskillDesignermakes the visualsHiggsfieldimagesVideo Editorcuts the clipsRemotionvideoSocial Leadfits each platformRepurposeskillPlannerschedules the weekNotioncalendar
0 of 6 moved into Claude
Click a seat to move it into Claude
Each person becomes a skill you keep and a tool you connect — reusable, on-brand, always on.
By the end of this workshop

One person. The whole agency.

You + ClaudeTHE OPERATORVoiceskillWritingskillRepurposeskillHiggsfieldimagesRemotionvideoNotioncalendar
Tap any capability to see what it replaces
By the end of this workshop this is yours: the work of six people, running through one. That’s what this workshop builds.
The map

Where we're going today.

1
Watch one prompt replace a week of work
Do it by hand first — then watch AI do it in one step.
Feel how long the manual way really takes, then see a single instruction collapse it into seconds. This is the idea behind everything in this workshop.
8:00
2
Run every post through four brand-safety checks
Tone, values, legal, off-limits — the four checks every post must pass.
Get comfortable with the tools you’ll use all week, then learn the four things every post has to get right — and practise catching problems in your own.
10:00
3
Teach Claude your brand voice
Define how it sounds, who it is, and what it must never say.
Set how your brand sounds, the personality behind it, and the words it should never use — so everything you make sounds like you.
12:30
4
Save it all as one reusable playbook
Combine the safety rules and your voice into your Brand Voice Profile.
Bring the safety rules and your brand voice together into the first part of your playbook — something you’ll reuse every day after this.
1:45
5
Connect your first two tools
Hook up Notion and Higgsfield so your agency is ready to run.
Connect the planning tool (Notion) and the image tool (Higgsfield) to Claude — your first two hires, live and ready for Day 2.
2:00
Before we start
How long to make a week of on-brand content —
by hand?

Be honest. Every platform. On-brand every single time. Hold that number.

The reveal

What if one prompt did all that?

MCP→prompt→no code
tap each piece ↓
Connect a tool's MCP to Claude, prompt it, and the work happens. The backbone of this whole week.
The engine · the whole week

So who's doing the work?

Here's the full engine you build in this workshop. Step through it — and tap any box for what it does.

MCP Youone prompt Claudeorchestrator ▶ Remotion — video ▦ Higgsfield — visuals ▤ Notion — calendar ♪ ElevenLabs — audio

One prompt in — all of this out.
Day 1 · safety
Brand Playbook
✓ safe
Day 2 · copy
A week of posts
LI
X
IG
TT
Day 2 · visuals
Image + video
Day 3 · ship
Calendar + analytics
Demo · the chat

What does that look like?

CClaudeon-brand mode
Write a launch post for our new serum — warm, on-brand, no medical claims, no hype.
Demo · under the hood

But how does Claude reach the tool?

Claudethe brain MCPthe adapter APIthe tool's door Tooldoes the work
MCP is the universal adapter; the API is each tool's door. You prompt — Claude speaks both.
Where you'll build

Where does this all happen?

Claude+
The engine. You drive the car — you don't study the pistons.
Your prompt lands here. Claude reads the intent, routes the work to each tool over MCP, and brings the result back — the brain of the whole operation.
Cowork+
A studio, not a text thread — it produces and keeps real files.
Where a chat window forgets, Cowork has a workspace: it runs the tools, writes the files, and saves what it makes so your work survives the whole week.
Artifacts+
The deliverables on the bench. Your BVP is a living document all week.
Every output Claude builds — your Brand Voice Profile, the posts, the calendar — lives as an artifact you can open, edit, and reuse, not a message you scroll past.
Analogy · the advanced bit

When do you stop reading the recipe?

RECIPEread every step internalise SKILLit just runs — and improves

The third time you make a dish, the steps are saved — you just cook, and tweak it a little better each time.

That saved, improvable recipe has a name in Claude.
Name it · skills

That's a skill.

a saved recipe you improve→a Claude skill
Build once Reuse Sharpen each time
Build it once, reuse it everywhere, sharpen it each use — you'll build one tomorrow.
Meet a tool · you'll use it Day 3

Remotion turns a prompt into video.

No timeline, no editor. You brief it, Claude drives it through its MCP, and a branded clip comes back. We meet it today — you build with it on Day 3.

brief→Claude + MCP→Remotion→video
tap each piece ↓
The same one move you just learned — pointed at video.
Feel it · the danger

How did this ship?

"Clinically proven to fix your skin in 7 days — guaranteed."
Tap the zone this breaks:
"Would your brand have caught it — before the public did?"
Analogy

What saves a fast car on a cliff road?

GUARDRAIL the edge

The guardrail doesn't slow the drive. It stops one bad turn from going off the edge.

AI is fast. Fast is exactly how off-brand content slips out.
Name it

Those guardrails have a name.

guardrails on the road→brand safety

And brand safety lives in four zones — every AI failure breaks one of them.

The four zones

Which zone did the post break?

Tone
The voice never drifts — no sarcasm where you're warm.
Values
Nothing that contradicts what the brand stands for.
Legal guardrails
No claims, guarantees, or outcomes you can't make.
Off-limits
Topics, claims, competitors you never touch.
Name the zone, write the rule. That's the whole game.
In the room · struggle window
8′ struggle / 4′ unlock

Can you break your own output?

Red-team = attack your own output to find what's unsafe before the public does.

Demo
Hassan walks the four zones on the slipped-through post.
You do it
Generate a post for your brand, then spot what broke and which zone.
Floor: 3 failures, by zoneStretch: write the catching ruleCatch-up: 1, clearly named
Reset

Breathe. Then we build the voice.

You've named the danger. After the break: how to make the AI sound unmistakably like you.

Analogy

How do you brief a new hire once?

BRIEFsaid once post email caption reel each inherits it

"Warm but never cheesy, no exclamation marks, never promise results." You say it once — they apply it to everything after.

That one-time briefing has a name in AI.
Name it

Brief it once. It sticks.

standing orders for a new hire→a system prompt

Set the rules once at the top, and every reply after obeys them — no reminding.

Engineer the voice

What's actually in a voice?

System prompt Persona Negatives → On-brand output
Voice isn't luck. It's a system prompt + a persona + a no-fly list.
See it · flip the voice

Same brief — whose brand?

Generic
Brand voice
Buy our new moisturizer now!! It's revolutionary and will totally transform your skin overnight, guaranteed!!!
system: warm, plainpersona: a trusted guideno "revolutionary"no false promisesno "!!!"
Negatives do the heavy lifting

What must it never say?

Most off-brand output isn't what the AI said — it's what it should never have said.

# negative constraints Never use "revolutionary" Never imply medical or guaranteed outcomes Never use more than one exclamation mark Never name a competitor
Your red-team failures? Each one becomes a line on this list.
Build · your reusable spec

Can a whole voice fit on one card?

BRAND VOICE PROFILE VOICE TONE ALWAYS NEVER

The Brand Voice Profile (BVP) — tone sliders, do/don't lists, and 3 sample prompts that already work. It lives as an artifact, and tomorrow becomes a skill.

The most-reused thing in the workshop. Everything downstream pulls from it.
Feel it · the BVP's sliders

Drag the tone. Watch it shift.

Our new serum supports your skin's natural barrier over time.
Tone: measured & careful
In the room

Build yours.

Demo
From real brand assets → a working BVP with the sliders and the no-fly list.
You do it
Build yours. Make it produce one on-brand post that passes the four zones.
Floor: sliders + do/don't + 1 promptStretch: banned-phrases + 3 adjectivesCatch-up: sliders + do/don't
Leave with → your BVP (artifact)
Integration · the deliverable

How do the pieces become one thing?

Four-zone rules Your BVP + PlaybookSection 1
Leave with → Playbook §1 (artifact)
Your first two tools

Connect your first two tools.

Youone prompt Claudethe brain MCPthe adapter Notionthe calendar Higgsfieldthe images
Notion holds your calendar; Higgsfield makes your images. Claude reaches both through their MCP — connect them once today, and they’re ready for the week ahead.
Demo · the home

Set up the calendar once.

MON
Launch post
TUE
Reel · hook
WED
Carousel
THU
Story set
FRI
Voiceover clip
SAT
+
SUN
+
Today you set up the structure — channels, days, status. Tomorrow you fill it with your real week, and Claude schedules it straight in for you.
Where we are

What did you build?

✓
Brand Voice Profile
✓
Playbook §1 — your safety rules + voice
✓
Notion + Higgsfield connected — your calendar and image tools, live
→
Tomorrow: a full week of content — produced and scheduled
Close

A brand AI can't break.

Tomorrow you stop re-typing your voice — you turn it into a skill, run one brief into seven days of content, and drop it straight into the calendar you just built.

Next → Day 2 · Full-Funnel Production
1 / 1
Break — how long?