How do you produce a full week of marketing content with AI?

Build a structured pipeline with 5 steps: keyword research and PAA collection, batch writing with a fixed content calendar and pillar assignments, image generation for blog headers and LinkedIn stats, scheduled publishing to the website, and saved drafts for manual channels. The key is research before writing and quality gates between each step.

Can Claude Code replace a marketing team?

No. Claude Code replaces production time, not strategy. The system follows a content calendar, voice guide, and research bank that a human created. It does not decide what to write about, what pillar to target, or which audience to serve. It executes the plan faster. The strategy, judgment, and engagement are still human.

How long does it take to set up a content production pipeline with AI?

The initial setup — content calendar, voice guide, research insights bank, pillar definitions, and the production skill — took approximately 2-3 weeks of iterative building. Once built, each weekly production run takes about 60 minutes. The setup cost is front-loaded but the production time compounds every week.

What tools do you need for AI-powered content production?

Claude Code (the AI engine), a website with a database (Supabase for blog storage), a publishing script (Node.js), Python for image generation (Playwright for rendering), and free tier accounts for newsletter (Beehiiv) and content tracking (Airtable). Total cost: approximately £80/month (Claude subscription). No paid content tools, design platforms, or scheduling services.

Does AI-produced content rank in search?

Yes, when the content is research-backed, properly structured for SEO (keyword-targeted titles, FAQ schema, internal links), and provides original data that competitors do not have. AI produces the draft. The research, data, and editorial judgment that make it rank come from the human who built the system and the original research it cites. ---

I Built a System That Produces a Full Week of Marketing Content in One Session

Q: Companion LinkedIn Post:

I used to spend 8-10 hours a week producing content. Sunday evenings planning. Monday mornings writing. Tuesday afternoons editing. By Wednesday I was behind. By Friday I had posted two of the three things I planned. The problem was never ideas. It was production overhead. Every blog post needed keyword research, PAA questions, competitor analysis, internal data, then writing, then images, then the LinkedIn companion, then the newsletter version. So I built a system. One session. One hour. The output: 4 blog posts, each with keyword-targeted titles, 5 FAQ questions from real search data, internal links, and external citations. 5 LinkedIn posts, each written as a story — not a data dump. 1 newsletter linked to Tuesday's blog. 9 branded images — 4 blog headers and 5 LinkedIn stat images. Everything scheduled or saved as ready-to-post drafts. The first week is live. 4 posts scheduled, going live at midnight UTC each day. LinkedIn posts scheduled for 5am. The part that surprised me: the hardest step was not getting AI to write. It was building the research layer underneath. A living insights bank with my own benchmark data. A keyword brief with real PAA questions. A voice guide strict enough that every post sounds like me, not like AI. The writing took 20 minutes. The research infrastructure took 3 weeks. That is the part most people skip. And it is the part that makes the output actually rank, actually resonate, and actually sound like it came from someone who does this work. The full system is open source. Blog post with the breakdown and tool links in comments.

I used to spend Sunday evenings planning content. Monday mornings writing. Tuesday afternoons editing. By Wednesday I was behind. By Friday I had posted two of the three things I planned.

The problem was never ideas. It was the production overhead. Every blog post needed keyword research, PAA questions, competitor analysis, internal data pulled from my own research, then the writing, then the image, then the LinkedIn companion, then the newsletter version. Each piece touched 6-7 steps before it was ready.

So I built a system that does all of it in one session.

What the System Produces

One command. One session. One hour. The output:

→ 4 blog posts (one per publishing day: Monday, Tuesday, Thursday, Friday) → 5 LinkedIn posts (daily, Mon-Fri, each in story format) → 1 newsletter (Tuesday anchor, linked to Tuesday's blog) → 4 blog header images (branded, pillar-tagged) → 5 LinkedIn stat images (data-driven, one per post) → 1 content brief with all keyword research, PAA questions, and competitor analysis

Everything saved locally. Blog posts scheduled with future publish dates. LinkedIn posts and newsletter saved as drafts ready for manual posting.

How It Works: 5 Steps

Step 1: Research

Before any writing happens, the system researches every blog post's keyword.

For each of the 4 posts, it: → Searches the primary keyword to find what competitors have published → Collects "People Also Ask" questions from search results (these become the FAQ section) → Pulls internal data from our Research Insights Bank (our own benchmark stats to cite) → Finds 3-5 external stats with source URLs for credibility → Identifies the competitor gap (what they cover vs what we have that they do not — usually original research data)

The output is a content brief: 4 topics, each with keywords, PAA questions, internal data, external data, competitor gap, and LinkedIn hook angle.

Quality gate: the brief is reviewed and approved before any writing starts.

Step 2: Write

With the research brief approved, the system writes all content in one batch.

Each blog post follows the same structure: → Hook (data-led or problem statement, first 2-3 sentences) → Body (problem → data → framework → steps → CTA, 1,000-1,500 words) → 3+ internal links to our resources (Scorecard, Playbook, Benchmark, Audit) → 1+ external link to a credible third-party source → 5 FAQ questions sourced from the PAA research → Companion LinkedIn post at the bottom (story format, not data dump)

The newsletter is a companion to Tuesday's blog — same topic, deeper POV, links to the blog for the full breakdown.

Wednesday's LinkedIn is a standalone POV post designed for engagement, no blog link.

Quality gate: all content is reviewed before image generation.

Step 3: Generate Images

For each blog post, the system generates a branded header image with the title, pillar tag, date, and author.

For each LinkedIn post, it generates a stat-based image: one big number, a caption, and a subtext line. These are the images that stop the scroll — a data point that makes someone pause and read.

9 images total: 4 blog headers + 5 LinkedIn stat images.

Step 4: Publish

Blog posts are published to the website with scheduled dates. Each post goes live at midnight UTC, ready for the 5am LinkedIn post that links to it.

The publishing script inserts the post into the database, uploads the header image, pushes metadata to the content tracker, and sets the publish date.

Step 5: Save Drafts

LinkedIn posts and the newsletter are saved as local files with the posting date in the filename. Since LinkedIn and Beehiiv do not have API publishing for my plan, these are manual — but the content is ready. Copy, paste, attach image, schedule for 5am.

What Made This Possible

Three things had to exist before this system could work:

1. A Research Insights Bank. Every research study I run writes its key findings to a single file. When the content system needs a stat, it reads from that file instead of searching the internet. The data is mine. The source links point to my own published research. I cite myself instead of quoting Gartner.

2. A fixed content calendar with pillar assignments. Monday is always Strategy. Tuesday is always Data (+ newsletter). Thursday is always Systems. Friday is always Leadership. Wednesday is a standalone POV. The system does not decide what type of content to produce. It follows the schedule.

3. Story format as a constraint, not a choice. Every LinkedIn post follows a discovery narrative structure. Start with a belief or a person. Introduce tension. Turn. Reveal. Reframe. Close with a question. The system does not produce data dumps. It produces stories backed by data. This is enforced as a writing rule, not a suggestion.

The Numbers From Week 1

The first full week produced:

→ 4 blog posts researched, written, and scheduled → 5 LinkedIn posts in story format with companion images → 1 newsletter with blog link → 9 branded images → 1 content brief with keyword research for every post

Total production time: approximately 60 minutes of active session time.

Previous production time for equivalent output (manual, spread across the week): approximately 8-10 hours.

What I Would Change

The system is not perfect. Three things I am iterating on:

Research step needs live search APIs. The keyword and PAA research currently uses web search tools. Connecting dedicated search APIs (Tavily, Brave Search) will make this faster and more reliable. Both are now connected and will be used from Week 2.

Thursday needs a research scan. The Systems pillar often involves scanning real websites and publishing the data. This means the Thursday blog sometimes depends on a batch scan that takes 30+ minutes. The scan needs to run earlier in the week so the data is ready.

LinkedIn scheduling is still manual. The blog posts publish automatically via the scheduling system. LinkedIn and the newsletter require copy-paste. This is a limitation of the free tier on both platforms, not a system design problem.

Try It Yourself

The content production skill, research tools, and image generators are all open source.

→ The weekly content skill defines the full pipeline → The blog image generator creates branded header images → The LinkedIn image generator creates stat-based images → The full tools page lists every tool we have built

The system runs on Claude Code with zero API costs beyond the Claude subscription. No paid content tools. No scheduling platforms. No design subscriptions.

What This Means for Enterprise Marketing Teams

Most enterprise marketing teams spend 60-70% of their time on production — writing, editing, formatting, scheduling, creating assets. The strategy gets 30 minutes on Monday morning. The rest of the week is execution.

A system like this inverts that ratio. Production takes an hour. The rest of the week is strategy, engagement, and analysis.

The question is not whether AI can produce marketing content. It is whether your team has built the infrastructure to make that production consistent, research-backed, and aligned to a strategy — or whether they are still prompting one piece at a time.