This is a practical walkthrough of how I stopped writing one-off posts and built a repeatable pipeline. If you want to skip building it yourself, there's a ready-made drop-in at the end — otherwise, copy the pieces below.
I used to write one LinkedIn post at a time.
Open a blank page. Stare at it. Give up. Write "How I learned X" for the 400th time. Post. Refresh for 90 minutes. Complain that reach is dead.
Then I did the obvious thing and stopped treating every post as original work. One blog post is actually six pieces of content — a thread, a LinkedIn post, a newsletter section, a YouTube intro, an Instagram carousel, and a Reddit comment — if you run them through a pipeline instead of rewriting from scratch every time.
Here is the pipeline. It's six stages, fully prompt-driven, and you can build every part of it yourself in an afternoon.
The 6-stage content pipeline
topic → outline → draft → edit → SEO → repurpose
↓ ↓ ↓ ↓ ↓ ↓
search structure voice polish rank multiply
data (×5 platforms)
Every stage is one prompt. Every stage consumes the previous stage's output. That's the whole trick.
Stage 1: Topic research (not brainstorm)
The mistake people make here is asking "give me 10 blog post ideas about X." You'll get the same 10 ideas everyone else is writing.
Instead, anchor on actual search behaviour:
You are a content strategist. My niche is [NICHE]. My audience is [AUDIENCE].
Give me 15 blog post topics that meet ALL of these criteria:
1. Answer a specific question people ask (not a "guide to" topic)
2. Have a clear point of view — not a neutral summary
3. Target a long-tail query (3+ words)
4. Can be written by someone with direct experience (no survey or study required)
Format each as:
- Title (question format)
- Search intent (informational / comparison / transactional)
- Point of view (the argument you'd make)
- Primary keyword
What changes: you get topics with an angle, not topics that will become Wikipedia-flavoured sludge. The "point of view" column is the one that separates decent from great.
Stage 2: Outline that is actually an argument
Don't let the model write a draft yet. Outline first, or you'll fight the whole draft trying to re-shape it.
You are a senior editor. Draft an outline for "[TITLE]" with this exact structure:
1. HOOK (1 paragraph): a concrete situation the reader recognises
2. STAKES (2 bullets): what it costs them if they keep doing the wrong thing
3. THESIS (1 sentence): the argument the whole post defends
4. BODY (4-6 sections): each section is a claim + evidence + example
5. OBJECTIONS (1 section): steelman the two strongest counterarguments
6. ACTION (1 paragraph): the single thing the reader should do next
Rules:
- Each body section must defend the thesis, not just expand it
- No "it depends" conclusions
- No "further reading" section
What changes: the outline is now a structured argument. The body sections defend a thesis instead of just "covering the topic." Drafts written from thesis-first outlines read like essays, not listicles.
Stage 3: First draft (in your voice, not Claude's)
The biggest problem with AI-written content is that it all reads the same. Solve this at the prompt level with a voice guide, not a post-edit.
Write a first draft of the outline below in my voice.
VOICE GUIDE:
- Short sentences. Then one long one that earns the space.
- Concrete specifics over abstractions. "$47" not "cheap." "3 weeks" not "recently."
- One em-dash per paragraph maximum.
- No em-dash sentences where a period works.
- No "it's important to note" or "let's dive in."
- No bullet lists in the intro or conclusion.
- Admit one thing that doesn't work. Real essays have friction.
OUTLINE:
[paste outline from stage 2]
The voice guide is the lever. Swap it out per platform/person and the same outline produces wildly different drafts. Keep a library of four voice guides: professional, casual, authoritative, contrarian. Use the one that matches the piece.
Stage 4: Two-pass edit
One pass is structural. One pass is sentence-level. They're different jobs and conflating them makes the model do neither well.
Structural pass:
You are a line editor doing a structural pass only. For the draft below:
1. Flag any paragraph that doesn't advance the thesis
2. Flag any section that should be moved, merged, or cut
3. Flag any claim that needs evidence but doesn't have it
4. Suggest the 1-2 biggest structural improvements
Do not rewrite. Mark up the draft with [FLAG: reason] inline.
DRAFT:
[paste]
Copy pass:
You are a copy editor. For the edited draft below:
1. Tighten every sentence that can be shorter
2. Cut every adverb that isn't earning its place
3. Replace abstract nouns with concrete ones
4. Kill every "in order to" (use "to")
5. Return the cleaned draft with no explanatory text
DRAFT:
[paste]
What changes: you stop getting generic "improved" drafts where the model changes whatever it feels like. Two narrow passes give you targeted, auditable edits.
Stage 5: SEO after, not before
SEO-first writing reads like SEO. Write the post for humans, then retrofit:
You are an SEO editor. The post below is finished. Do NOT rewrite the body.
Your job is exactly three things:
1. Title + meta description optimised for "[PRIMARY KEYWORD]" — two versions each
2. FAQ section (4-6 questions) using real People Also Ask phrasing — short direct answers
3. Internal linking suggestions: 3 topics this post should link TO and 3 topics this post should be linked FROM
POST:
[paste]
The FAQ block is the cheat code. Google surfaces them as featured snippets and they're genuinely useful — not hacky.
Stage 6: Repurpose (one piece → six)
This is where the compounding happens. The same draft becomes:
You are a content distributor. Turn the post below into SIX pieces:
1. TWITTER THREAD (7-10 tweets): hook tweet is a specific claim with a number, each tweet stands alone, last tweet CTAs to the full post.
2. LINKEDIN POST (150-250 words): first-person story framing, lesson at the end, no link in body (link in first comment).
3. NEWSLETTER SECTION (200 words): "here's what I wrote this week" framing, link with context.
4. YOUTUBE SHORT SCRIPT (45 seconds, ~130 words): hook in 3 seconds, one insight, CTA.
5. INSTAGRAM CAROUSEL (6 slides, max 15 words each): slide 1 hook, slides 2-5 one idea each, slide 6 CTA.
6. REDDIT COMMENT (3 paragraphs): written as a response to a hypothetical question — no self-promotion, but signals expertise.
POST:
[paste]
Rules: each piece must be natively formatted for the platform. Do NOT cross-post one format to another.
The key word is natively. A Twitter thread that's a LinkedIn post copy-pasted is dead on arrival. Each platform has its own physics and the prompt has to force the model to respect them.
Running it end-to-end
Six stages. Six prompts. Paste output from one into the next. That's the whole pipeline.
In practice, this takes me about 90 minutes for a 1500-word post plus six platform variants. Before, it took 3-4 hours for the post alone and I distributed maybe 2 of the 6 variants.
The delta isn't the AI — it's the pipeline. Running the prompts in isolation gets you mediocre output at each stage. Chaining them, with each stage constrained to one narrow job, is what makes the system work.
Where it breaks
Honest version: this pipeline doesn't work if you don't have a point of view. The "thesis" in stage 2 is load-bearing. If you don't know what you actually think about the topic, every downstream stage produces professional-sounding sludge.
The pipeline amplifies whatever you put in. Strong angle → six strong pieces. No angle → six flavours of nothing.
Fix that first. Everything else is mechanical.
Going further
If you want the full pipeline with all six prompt chains, four pre-built brand voice guides, five platform playbooks (Twitter/X, LinkedIn, YouTube, Newsletter, Blog), three content calendars, and two Zapier automation recipes packaged and ready to drop into your workflow, I assembled the whole thing into The Content Factory — the same system I use daily, $39, lifetime access.
Otherwise, build it yourself from the prompts above. The framework matters more than the packaging.
Top comments (0)