AI Content Pipelines: From Blog Idea to SEO-Ready Draft in 30 Minutes
Full disclosure before we start: this post about an AI content pipeline blog SEO workflow was written using a version of the same pipeline I'm about to describe. That's the meta-irony, and it's also the proof. The first draft came out of four chained skills in 28 minutes. I then spent another 90 minutes rewriting the parts that smelled like AI, fact-checking the time estimates against my own logs, and arguing with myself about whether the lead should open with the disclosure or bury it. So: 30 minutes to a draft, two hours to a post that's actually worth reading. Anyone selling you "publish in 30 minutes" is selling slop.
Table of Contents
Why most AI content pipelines produce slop
Search "AI content pipeline" and you'll get fifty Notion templates that all do the same three things: feed a topic to ChatGPT, ask for an outline, ask for a draft. That's not a pipeline — it's one prompt with extra steps. The output reads like every other AI blog post on the internet because it was written like every other AI blog post on the internet.
Real pipelines split the work into stages where each stage has a specific job and a specific constraint. Brainstorming wants quantity and variance. Writing wants structure and voice. SEO wants discipline. Humanizing wants taste. If you ask one model in one prompt to do all four, you get the median of all four — which is exactly the bland, faintly nautical, em-dash-laden voice that makes readers bounce in two seconds.
The pipeline below uses four separate skills, one per stage, each tuned for its job. The whole thing runs on Claude Code or any agent that reads SKILL.md files. If you're new to skills as a primitive, the prompts-to-agent-skills walkthrough covers the mental model. If you've already drunk the kool-aid, skip ahead.
The 30-minute AI content pipeline, step by step
Four skills, four stages, one chained workflow. I'll give you the elapsed time for each step from my last run, which produced a 1,400-word post on email deliverability for a SaaS client. Your mileage will vary based on topic complexity and how picky you are. I'm picky.
Step 1: content-brainstorm — topics and angles (4 minutes)
What it does: takes a vague theme ("we should write more about email") and produces 15–25 specific post ideas, each with a target keyword, a rough angle, a reader persona, and a one-line hook. Critically, it scores them on a "boring vs. ownable" axis so you can skip the obvious takes.
Why a content marketer needs it: the bottleneck on most content calendars isn't writing speed — it's deciding what to write. You stare at a blank Asana column for 20 minutes, pick the safest idea, and ship something nobody reads. content-brainstorm gives you 20 ideas in 4 minutes, ranked, so you can pick the one with actual conviction behind it.
Last run: I gave it "email deliverability for B2B SaaS" and got back 22 ideas. Two were obvious ("DMARC explained" — already covered by every vendor blog). Three were ownable: a teardown of why Postmark's bounce rates are lower than Sendgrid's at the same volume, a piece on what changed for cold outbound after Gmail's Feb 2024 sender rules, and a "we got blacklisted" postmortem. I picked the cold-outbound one.
Step 2: content-writer — outline and draft (12 minutes)
What it does: takes the chosen idea, target keyword, and reader persona, then produces a structured outline (H1, 5–8 H2s, lead paragraph, FAQ section) followed by a full first draft. The skill's job isn't to write something publishable — it's to write something shapeable. Good bones, decent voice, gaps clearly marked where you need to add a real example or an opinion.
Why a content marketer needs it: first drafts are the worst part of writing. If you can offload the "stare at the blinking cursor" stage to a skill that produces something 60% of the way there, you spend your energy on the 40% that actually matters: the angle, the anecdotes, the data, the take.
Last run: 12 minutes. The skill produced a 1,200-word draft with seven section headers and four bracketed gaps that said things like [INSERT REAL EXAMPLE OF A COLD OUTBOUND CAMPAIGN AFFECTED BY THE FEB 2024 RULES]. Those brackets are the feature, not the bug. They're where I stop being a button-pusher and start being a writer.
If you care about voice consistency across a whole content program, pair this skill with the brand voice rewriting workflow — it's the cleanest way I've found to keep a portfolio of AI-assisted posts sounding like one writer wrote them.
Step 3: seo-optimizer — keyword discipline and structure (7 minutes)
What it does: takes the draft and the target keyword, then audits keyword placement (H1, lead, ≥1 H2, FAQ first question), checks heading hierarchy, flags missing alt text, suggests internal links from a list of your existing posts, and rewrites the meta title and description to fit Google's pixel limits. It does not stuff keywords. If you ask it to, it pushes back. If you want to feed it a competitor's outline as input, paste their URL into our free web scraper first — you'll get every heading, every internal link, and the JSON-LD they ship in thirty seconds.
Why a content marketer needs it: SEO is the most rule-bound part of writing, which means it's the easiest part to delegate to a skill. The rules don't change post-to-post. You don't need creativity here — you need a checklist that gets run consistently every time. Humans forget the checklist. Skills don't.
Last run: 7 minutes. The skill caught that I'd used "email deliverability" 11 times but the real long-tail target was "cold outbound deliverability 2024" — which appeared zero times in the draft. It rewrote two H2s, three sentences, and the meta description. Search Console six weeks later showed the post ranking page-one for the corrected keyword. The original would've ranked nowhere.
Step 4: content-rewriter — humanize and de-slop (5 minutes for the pass, longer if you're honest)
What it does: reads the SEO-optimized draft and rewrites the parts that smell like AI. It targets the usual tells: tricolons, em-dash overuse, "in today's fast-paced," "moreover," empty hype words, that weird rhythmic cadence where every paragraph is exactly three sentences. It also varies sentence length deliberately. Short. Then a longer one with a clause that breaks the pattern. Then short again.
Why a content marketer needs it: the SEO-optimized draft will rank, but readers will still bounce if it sounds like a robot. content-rewriter is the layer between "Google likes it" and "humans actually finish reading it." This is the step most pipelines skip. It's why most AI content pipelines produce slop.
Honest caveat: 5 minutes is the skill's pass. Then you read the output and rewrite the parts the skill missed, because no skill catches everything. On the email deliverability post, the skill flagged 14 issues and fixed 11. I caught three more myself — including a paragraph that was technically correct but read like a LinkedIn post written by someone who'd just discovered the word "leverage." That part I rewrote from scratch.
Where the 30 minutes actually goes
Adding it up from the email deliverability run:
- content-brainstorm: 4 minutes (mostly Claude thinking; I picked from the list in about 90 seconds).
- content-writer: 12 minutes (longest stage, because a 1,200-word draft takes time to generate even on a fast model).
- seo-optimizer: 7 minutes (audit + rewrites of two H2s).
- content-rewriter: 5 minutes for the automated pass.
- Total: 28 minutes, draft in hand.
Then the part nobody on Twitter mentions: another 60–120 minutes of human work to make it actually good. Filling the bracketed gaps with real examples. Adding the data point that proves the angle. Cutting two paragraphs that didn't earn their place. Writing the lead three times. Reading it out loud and fixing the line that sounded weird.
The pipeline isn't a "publish in 30 minutes" magic button. It's a "get to the interesting part of writing in 30 minutes instead of 3 hours" magic button. That distinction is the entire point.
If you're using this for inbound social distribution after publish, the Reddit growth playbook has the companion workflow for turning the finished post into a thread that doesn't get downvoted on sight. The two pipelines together are most of what a solo content function needs.
What this pipeline won't fix
Three honest limitations, because pretending otherwise is exactly the kind of slop the pipeline is supposed to prevent:
- It can't generate opinions you don't have. The brainstorm skill will surface ownable angles, but the actual conviction has to come from you. If you've never run a cold outbound campaign, the post on cold outbound deliverability will read like research, not experience. Readers can tell.
- It can't do real research. content-writer will happily make up plausible-sounding statistics. The skill is good at structure and voice; it's not a research agent. Numbers, quotes, and named examples need to come from your own data, your own customers, or a deliberate research step before the pipeline runs.
- It won't fix bad topic strategy. If your content calendar is built on keywords nobody searches for, the pipeline will produce beautifully optimized drafts about topics nobody cares about. The pipeline accelerates execution. It doesn't fix the upstream decision about what to execute on.
None of those are reasons to skip the pipeline. They're reasons to install it with eyes open and to keep a human in the loop where humans add value.
FAQ
Does this AI content pipeline blog SEO workflow work for non-blog content like landing pages or emails?
Partially. content-brainstorm and content-rewriter port over cleanly. content-writer is tuned for long-form blog structure (H1 / H2 / FAQ), so it'll fight you on a 200-word landing page. seo-optimizer is blog-specific. For landing pages and emails, run brainstorm and rewriter from this stack and pair them with copy-specific skills.
How does this compare to using just one tool like Jasper or Copy.ai?
Single tools optimize for "anyone can use this in five minutes." Skill-based pipelines optimize for "someone who cares about output quality can build a workflow that compounds." If you're a marketer producing one post a month, a single tool is fine. If you're shipping weekly or running a content program for clients, the per-stage approach gives you better output and lets you swap individual stages without rebuilding the whole thing.
Can I run this pipeline in something other than Claude Code?
Yes. The four skills are Markdown files. They run on any agent that reads SKILL.md format — Claude Code, Cursor with light adaptation, and most of the agentic IDEs. If you're orchestrating across multiple skills programmatically, see the agent skills workflow guide for chaining patterns.
What's the cost per post in API tokens?
On my email deliverability run, the four-stage pipeline used roughly 35,000 input tokens and 8,000 output tokens across the chain. At current Claude Sonnet pricing that's well under a dollar per post. Cheaper than the coffee you drink while it runs.
Won't Google penalize AI-written content?
Google's stated policy is that they penalize unhelpful content, regardless of how it was made. The pipeline above is specifically designed to produce helpful content — the rewriter step exists to remove the AI-pattern signals that correlate with unhelpful output. Posts I've published using this exact pipeline have ranked page-one for competitive keywords. The slop concern is real; the AI-origin concern is mostly noise.
How do I know if my output is still slop?
Read it out loud. If you sound like a LinkedIn motivational poster, it's slop. If three paragraphs in a row are the same length, it's slop. If you've used "leverage" as a verb, it's slop. The rewriter skill catches most of this, but the final taste check is yours.
The bottom line
An AI content pipeline isn't a shortcut to publish-ready content — it's a shortcut to the part of writing that actually deserves your time. The four skills above (content-brainstorm, content-writer, seo-optimizer, content-rewriter) handle the mechanical 80% in about half an hour. You handle the creative 20% that makes the post worth reading. That ratio is the right one.
If you want to start with one skill instead of four, install content-rewriter first and run it over your last published post. The diff will tell you exactly how much slop your current process is shipping. That's usually all the convincing anyone needs.
→ Browse all marketing skills · Request a skill that doesn't exist yet
