Build Your Content Agent

Create a CONTENT.md that tells an agent exactly how to produce content for your business, then wire it up to generate blog posts, social copy, and SEO drafts — continuously, without you in the loop.

Why This Matters

Most founders who try AI for content get mediocre output because they're prompting from scratch every time. No shared context. No brand memory. No style guardrails.

A content agent is different. It knows your brand, your ICP, your keywords, your editorial standards — because you wrote them down in CONTENT.md. Every piece it generates starts from that foundation.

The goal isn't one great blog post. It's a content system that runs without you having to think about it.

What You're Building

CONTENT.md — The editorial brain:

Keywords and topics you target
Content types you publish and their format specs
Approval workflow (what gets human review vs. auto-publish)
SEO rules (word count, heading structure, internal linking)
What a "great" post looks like for your audience
Content calendar logic (cadence, seasonal angles, evergreen vs. timely)

A working content agent — configured in Claude Projects, Claude Code, or Paperclip — that reads CONTENT.md and produces publication-ready drafts.

Implementation Tiers

Tier 1: Claude Projects (15 minutes)

Add CONTENT.md to your Claude Project knowledge alongside COMPANY.md and MARKETING.md. Use it as context for a dedicated content conversation.

Step 1: Generate CONTENT.md

Based on my COMPANY.md and MARKETING.md, help me write CONTENT.md — the editorial playbook for an AI content agent.

Include:
1. Primary keywords — 10-15 keywords and phrases we target, with monthly search volume if known
2. Content types — blog posts, Twitter threads, LinkedIn posts, email newsletters (format specs for each: length, structure, tone)
3. Blog post template — exact structure we follow (intro format, H2 structure, CTA, meta description format)
4. Approval workflow — which content types auto-publish vs. require human review
5. SEO rules — minimum word count, keyword density guidance, internal linking rules, image alt text
6. Quality bar — what makes a piece "good enough to publish" vs. "needs revision"
7. What we never write about — topics, claims, or formats that don't fit our brand

My content goals:
- [e.g., rank for long-tail keywords around [topic]]
- [e.g., build Twitter following in [niche]]
- [e.g., nurture email list with weekly tips]

Step 2: Run your first content brief

Using CONTENT.md, write a full blog post brief for:
Topic: [your topic]
Target keyword: [keyword]

Brief should include:
- Working title (3 options)
- Target reader (specific persona from COMPANY.md)
- Angle / hook
- Outline (H2s with 2-sentence description of each)
- Key stat or proof point to lead with
- CTA at the end
- Meta description (155 chars)

Step 3: Generate the full post

Write the full blog post from this brief. Follow the format in CONTENT.md exactly.
Aim for [word count] words. Include internal links to [your other posts/pages] where relevant.

Review, edit, publish. You've got your first agent-drafted post.

Tier 2: Claude Code (30 minutes)

Use Claude Code with your Company OS files to produce an entire content calendar and batch-generate drafts locally.

Step 1: Scaffold the content system

Create context/CONTENT.md using the prompt above. Then:

mkdir -p content/blog content/social content/email

Step 2: Generate a content calendar

In Claude Code:

Read COMPANY.md, MARKETING.md, and CONTENT.md.

Generate a 30-day content calendar with:
- 8 blog post topics (with target keyword and angle)
- 20 Twitter/X threads (5 per week, varied formats)
- 4 email newsletter topics (one per week)

Format as a markdown table with: Date | Type | Topic | Keyword/Angle | Status

Save to content/calendar.md

Step 3: Batch-generate drafts

From content/calendar.md, write the first 3 blog posts in full.
Follow all rules in CONTENT.md. Save each to content/blog/[slug].md.
Include meta description and internal link suggestions in each file.

You now have 3 publication-ready drafts in under 30 minutes.

Tier 3: Paperclip (1 hour)

A Paperclip content agent wakes up on a schedule, pulls the next item from the content calendar, generates the draft, and creates a task for your review — without you ever opening a prompt.

Step 1: Create your CONTENT.md and wire it to the agent

Add to your CMO or Content Agent's AGENTS.md:

## Content Agent Instructions

Before any content task, read:
- `COMPANY.md` — product, ICP, voice
- `CONTENT.md` — editorial standards, keyword targets, format specs

For each content task:
1. Identify the target keyword and content type from the task description
2. Write a brief (title options, outline, hook, CTA)
3. Generate the full draft following CONTENT.md format
4. Post the draft as a comment on the task and mark in_review for human approval
5. Never auto-publish — all content requires board review before publication

Step 2: Create your content calendar as Paperclip tasks

POST /api/companies/:companyId/issues
{
  "title": "Blog post: [topic]",
  "description": "Target keyword: [keyword]\nAngle: [angle]",
  "assigneeAgentId": "[content-agent-id]",
  "status": "todo"
}

Or create them in bulk from a CSV via script.

Step 3: The agent takes it from here

When the task is assigned, the content agent:

Reads COMPANY.md + CONTENT.md
Drafts the full post
Posts it as a task comment
Sets status to in_review
You review, approve, and it gets published

One human reviewing and approving. Zero humans writing first drafts.

Deliverable

CONTENT.md — complete editorial playbook
At least 3 published or publication-ready content drafts
A content calendar for the next 30 days

See the CONTENT.md template for a starting point.

What's Next

With content production running, move to Build Your Outreach Agent — where you set up an agent that monitors communities, surfaces opportunities, and drafts outreach messages for your review.