Part 2 of 7

Part 2 — The Playbook: Rules and Standards

ai claude cowork marketing rules brand standards cowork-rules

Created: April 17, 2026 | Modified: April 19, 2026

Cowork Features

Introduced: Rules (.claude/rules/)

This is Part 2 of a 7-part series on building your AI VP of Marketing with Claude Cowork. Previous: Part 1 — The Hire | Next: Part 3 — Skills

Why a Playbook

Your VP has the employee handbook. They know the business, the audience, the quarterly goals, the accountability zones. Now you write the policies.

CLAUDE.md answers "what do you need to know about us?" Rules answer a different question: "how must you behave?"

Rules live in a .claude/rules/ directory inside your project. Each rule is a markdown file containing instructions that fire on every single interaction. Not most interactions. Every one. When your VP drafts a blog post, the rules are active. When it writes an email subject line, the rules are active. When it outlines a social calendar, the rules are active. They are constraints, not suggestions.

CLAUDE.md is the employee handbook your VP reads on day one. Rules are the policies posted on the wall that govern every decision. Your VP references the handbook when relevant. The policies apply whether they are relevant or not — they shape behavior at a structural level, before creative work begins.

Rules vs. CLAUDE.md vs. Memory

CLAUDE.md provides context that your VP draws on when needed. Rules provide constraints that your VP follows unconditionally. Memory (from Part 1) is AI-curated and global to the Project — what your VP has picked up about you. Rules are human-authored and file-scoped — what your VP must do. All three load into every conversation, but each one answers a different question: what do you need to know, what have you picked up, and how must you behave.

Your Rules Directory

Your Cowork Project folder grows its first subdirectory beside CLAUDE.md:

your-cowork-project/
├── CLAUDE.md
└── .claude/
    └── rules/
        └── process-rules.md

This Part fills in three more rule files beside process-rules.md. Create four files under .claude/rules/: brand-voice.md, content-standards.md, process-rules.md, and approval-criteria.md. Each governs a different dimension of your VP's output. You can ask Cowork to create this structure for you:

Create a .claude/rules/ directory with four empty files:
brand-voice.md, content-standards.md, process-rules.md, and approval-criteria.md

Under the hood — Rules

Rules are additional instruction files that stack on top of CLAUDE.md without overwriting any line.

What file. Rule files live at ./.claude/rules/*.md under your Cowork Project folder on disk.
When written. You write Rules by hand when you save a rule file — Cowork never writes here.
What format. Plain markdown, with one file per rule domain like tone, format, brand, or approval.
How to inspect. Open any file in ./.claude/rules/ in a text editor, or browse the folder directly.
How to undo. Delete or edit the rule file on disk — the next conversation loads what remains.

Gotcha. Stacking is the whole point and also the whole risk. If two Rules disagree, your VP will try to honor both and produce something that satisfies neither. When you add a Rule, re-read the Rules already in place and resolve any conflicts deliberately.

Now fill them in.

Brand Voice Rules

Every piece of marketing your VP produces passes through this filter. Get it wrong and your content sounds generic. Get it right and people cannot tell your VP wrote it.

The example below uses a B2B SaaS company called RouteLine that sells logistics optimization software.

# Brand Voice

## Tone
Direct, confident, grounded, warm, specific.

We sound like a smart colleague explaining something clearly over coffee --
not a salesperson reading a script, and not an academic writing a paper.

## Personality Traits
- We are practical before clever. Utility beats wit.
- We respect our audience's time. Short beats long. Clear beats impressive.
- We show, then tell. Data and examples come before claims.
- We are honest about limitations. "This works best for X" beats "this works for everyone."

## Vocabulary

### Words We Use
- ship, build, fix, track, measure
- customers (not "users" or "end users")
- plain numbers ("40% faster") over vague scale ("significantly faster")

### Words We Never Use
- synergy, leverage (as a verb), disrupt, revolutionary
- game-changing, best-in-class, world-class, cutting-edge
- empower, unlock, unleash, supercharge
- "we are excited to announce" or any variation

## Voice in Action

**Wrong:** "We are excited to announce our revolutionary new feature that empowers logistics teams to unlock unprecedented efficiency gains."

**Right:** "Route optimization now runs in 3 seconds instead of 45. Your dispatch team gets their morning routes before coffee is ready."

**Wrong:** "RouteLine leverages cutting-edge AI to deliver best-in-class logistics solutions."

**Right:** "RouteLine uses route history and traffic patterns to cut delivery times. Most customers see a 20-30% improvement in the first month."

## Sentence Structure
- Lead with the benefit or the fact. Never lead with "we."
- Vary sentence length. Follow a long sentence with a short one.
- Active voice always. "You save 3 hours per week" not "3 hours per week are saved."

You probably do not sell logistics software. The structure matters more than the specifics. To generate a brand-voice.md tailored to your business, run the Brand voice interview in the companion Prompts panel. It walks your VP through your CLAUDE.md, asks the voice questions one at a time, and writes the file to ./.claude/rules/brand-voice.md on your approval.

After Cowork generates the file, read the "Voice in Action" examples out loud. If the "right" examples do not sound like something you would actually say to a customer, revise them. These examples are the strongest signal your VP has about your voice. They do more work than the adjectives.

Do not list 20 banned words and call it a voice guide. A wall of prohibitions teaches your VP what to avoid but not what to aim for. The positive examples — the "right" versions — are where voice actually lives. Spend more time on those than on the banned list.

Content Standards

Brand voice governs how your VP writes. Content standards govern what the writing looks like structurally. Different content types have different requirements, and your VP needs those boundaries stated explicitly. A short excerpt of .claude/rules/content-standards.md:

# Content Standards

## General Formatting
- Use sentence case for all headings ("How to reduce delivery times" not "How To Reduce Delivery Times")
- One idea per paragraph. Maximum 4 sentences per paragraph.
- Use subheadings (H2, H3) every 200-300 words in long-form content.
- Bulleted lists for 3+ parallel items. Never a bulleted list with only 2 items.
- Bold key phrases sparingly — no more than 2 bold phrases per section.

## Length Guidelines by Content Type

### Blog Posts
- Target: 800-1200 words
- Structure: Hook (2-3 sentences) → Problem → Solution → Evidence → CTA
- Must include at least one specific data point, example, or case reference
- Minimum 3 subheadings

### Email Campaigns
- Subject line: 6-10 words, no punctuation tricks
- Body: 150-300 words
- One CTA per email. One. Not two, not three.
- Preview text must be written intentionally, not left to auto-generate from body

### Social Media
- LinkedIn: 150-300 words for thought leadership, 50-100 words for announcements
- Twitter/X: under 250 characters for single posts, up to 4 tweets for threads
- No hashtags in the first sentence. Ever.

### Landing Pages
- Headline: under 10 words
- Subheadline: one sentence that expands the headline
- Every landing page ends with a clear, single-action CTA

## Content Requirements (All Types)
- Every piece of content must include a call to action
- Lead with the value to the reader, not a description of us
- No unverified statistics. If a number is used, it must be sourced or clearly labeled as internal data.
- No filler introductions. First sentence must earn the reader's attention.

Run the Content standards interview in the Prompts panel for the full version that tailors this file to the channels you actually publish on.

These standards should match your actual needs, not some theoretical ideal. If you only post on LinkedIn and send email newsletters, delete the Twitter and landing page sections. Unused rules add noise. Your VP performs better with fewer, relevant rules than with a comprehensive rulebook it has to sift through.

Process Rules

In Part 1, you built an accountability framework with review checklists and approval workflows. Process rules encode that framework so your VP follows it automatically, without you restating it each session.

The core structure of .claude/rules/process-rules.md:

# Process Rules

## Workflow Sequence
All content follows this sequence. Do not skip steps.

1. **Research**: Identify the target audience segment, key message, and supporting evidence before writing anything.
2. **Outline**: Present a structured outline for approval before drafting. The outline must include the headline, subheadings, key points per section, and the CTA.
3. **Draft**: Write the full draft following brand voice and content standards.
4. **Self-Review**: Run the Quality Checklist (below) against the draft. Fix any failures before presenting the draft.
5. **Present for Review**: Show the draft with the completed checklist. Flag any items that are borderline or need my input.

## Quality Checklist
Before presenting any draft, verify:

- [ ] Brand voice matches tone descriptors (direct, confident, grounded, warm, specific)
- [ ] No banned vocabulary used
- [ ] Length falls within guidelines for this content type
- [ ] CTA is present and specific (not "learn more" or "get in touch" without context)
- [ ] First sentence earns attention — no throat-clearing
- [ ] Active voice throughout — no passive constructions
- [ ] All statistics sourced or labeled as internal data
- [ ] Heading structure follows content standards
- [ ] Paragraph length within limits (4 sentences max)

## Channel-Specific Requirements

### Email
- Subject line must be presented separately, before the body
- Preview text must be written explicitly

### Social Media
- Hashtags listed separately from post body for easy editing
- Platform specified in the draft header

### Blog
- SEO title and meta description must be provided with the draft
- Featured image direction or description included

## When Stuck
If a task is ambiguous, missing information, or conflicts with existing rules:
- Ask for clarification before proceeding
- Do not guess at target audience, statistics, or claims
- State what information you need and why

Connecting to Part 1

The Quality Checklist is the codified version of the accountability framework from Part 1. There, you learned why review checklists matter and how to hold your VP to a standard. Here, that standard becomes a rule that runs on every task. Your VP will self-review against this checklist before showing you any draft.

Run the Process rules interview in the Prompts panel to tailor the workflow to the content types you actually produce.

A common mistake is writing process rules that describe an ideal workflow you do not actually follow. If you never write outlines for social media posts, do not require them. Rules your VP follows but you ignore create friction. Match the process to how you actually want to work.

Approval Criteria

The final rule file defines what "done" looks like. This is the gate between draft and finished work. Without it, your VP has no way to judge whether something is ready for your review or still needs iteration.

The structure of .claude/rules/approval-criteria.md:

# Approval Criteria

## Definition of Done
A piece of content is ready for my review when ALL of the following are true:

### Quality Gates
1. **Checklist Complete**: Every item in the Quality Checklist (process-rules.md) passes. No exceptions, no "close enough."
2. **Voice Match**: The draft reads like our brand, not like generic marketing copy. Test: could a competitor publish this without changing anything? If yes, it is not specific enough.
3. **Claims Verified**: Every statistic, data point, or factual claim is either sourced or explicitly marked as "internal estimate" or "to be verified."
4. **CTA Clarity**: The call to action tells the reader exactly what to do next and what they get from doing it. "Schedule a demo to see route optimization on your data" passes. "Learn more" does not.
5. **Audience Fit**: The content speaks to the specific audience segment identified in the research phase. A post targeting operations managers should not read like a post targeting C-suite executives.

### Presentation Requirements
When presenting a draft for review, include:
- The completed Quality Checklist with pass/fail for each item
- The target audience segment and channel
- The specific CTA
- Any items flagged as borderline or needing my input
- Suggested revisions if any checklist items required compromise

### Rejection Triggers
If any of the following are present, the draft is not ready. Revise before presenting:
- Passive voice in more than one sentence
- Any word from the banned vocabulary list
- Missing CTA
- Opening sentence that describes the company instead of addressing the reader
- Unverified statistic without a source label
- Content length outside the specified range for its type

## Escalation
If a requirement conflicts with another requirement (e.g., length limit versus thoroughness), flag the conflict and present two options with a recommendation. Do not silently choose one over the other.

This file works hand-in-hand with the process rules. Your VP runs the workflow, self-reviews against the checklist, and then checks the approval criteria before presenting work to you. Three layers of quality control, all automated, all running before you see a single draft.

Run the Approval criteria interview in the Prompts panel for the tailored version.

The "Voice Match" test in gate 2 is the most valuable check in this file. Generic content is the default failure mode of AI-generated marketing. If your VP can swap your company name for a competitor's and the copy still works, it has not done its job. This gate catches that.

How the VP Uses Rules

On every marketing turn, Cowork loads CLAUDE.md, Memory, and each file in .claude/rules/ as unconditional instructions. No manual trigger. The files are read, the constraints are applied, the draft comes back shaped.

Precedence is worth stating plainly. CLAUDE.md is context your VP reasons with. Rules are constraints your VP does not override. When Rules disagree with each other, your VP will try to honor both — which is why the Gotcha box above matters. If brand-voice.md says "never use the word unlock" and content-standards.md says "every email must include an unlocking value statement," one of them will lose and neither of you will know which. Resolve conflicts in the file, not in the conversation.

Putting It All Together

With all four rule files in place, your .claude/rules/ directory looks like this. This zoom is Part-2-local; the series-wide tree view grows in later Parts.

.claude/
└── rules/
    ├── brand-voice.md        # How we sound
    ├── content-standards.md  # What our content looks like
    ├── process-rules.md      # How we work
    └── approval-criteria.md  # When work is done

These four files, combined with the CLAUDE.md briefing document from Part 1, give your VP everything it needs to produce consistent, on-brand marketing work. CLAUDE.md provides the business knowledge. Rules provide the behavioral constraints. Together, they turn a general-purpose AI into your VP.

Test the setup now. Give your VP a task and watch the rules in action:

Write a LinkedIn post announcing that we just reduced average route
optimization time from 45 seconds to 3 seconds. Target audience:
operations managers at mid-market logistics companies.

Your VP should follow the workflow sequence from process rules (research, outline, draft, self-review), write in the brand voice, meet the LinkedIn length guidelines from content standards, and present the draft with a completed quality checklist. If any of those steps are missing, check the corresponding rule file.

Off-Ramp 1: What You Have Built

What you have built: A working AI VP of Marketing with full business context, an accountability framework, and codified brand standards. Your VP knows your voice, your audience, and your quality bar.

What is ahead: Parts 3-5 build reusable Skills, Agents, and the pipeline that turns this foundation into automated content production. Worth doing when you are ready — but what you have now is already a powerful marketing partner.

This is a real stopping point. Plenty of small businesses run effective marketing with exactly this setup: a well-briefed VP with clear standards. You hand it tasks, it produces drafts that sound like you, you review and publish. No automation, no pipelines.

If you stop here, you have something most businesses do not — a marketing partner that remembers your voice, follows your process, and holds itself to your standards on every single task.

What just changed

You wrote four files into .claude/rules/ — brand-voice.md, content-standards.md, process-rules.md, and approval-criteria.md — beside the ./CLAUDE.md from Part 1. Cowork loads them on every marketing turn, so voice and approval bar apply automatically. Your VP's rulebook is on the wall.

What is Next

Your VP has context, strategy, and standards. Now it is time to give them their first reusable tool.

In Part 3, you build a Content Brief Generator as a Cowork Skill — a saved, repeatable workflow your VP runs on demand. Skills load these Rules on every invocation. Instead of writing a new prompt every time you need a content brief, you trigger the Skill and get a consistent, standards-compliant brief in seconds. It is the first piece of real automation in your marketing operation.

This is Part 2 of 7 in the Your AI VP of Marketing series. Previous: Part 1 — The Hire | Next: Part 3 — Skills

Part 2 of 7