Claude Code SEO Automation: How I Publish a Ranking Post Every Weekday (No API)

Most articles about Claude Code SEO automation are lists. "12 workflows." "7 ways." Each one is a clever prompt you run by hand, get a result, and copy somewhere else. That is not what I mean when I say Claude Code runs my SEO.

I run one engine across three blogs. Every weekday it picks a topic, researches the pages already ranking for that topic, writes something built to beat them, and publishes the finished post to a live site — no human in the loop until the safety gate at the end. It reviews Google rankings once a week and refills its own content calendar. I wrote this post to show you the whole machine, not another prompt to paste.

I am a solo founder. I have around 14 launched products on Supabase and Vercel and three SEO blogs feeding them. I do not have a content team. The only way that math works is if the writing, the research, and the publishing happen without me sitting there. So that is what I built, and it is what this article is actually about.

What "Claude Code SEO automation" really means (it is not a prompt)

Claude Code is Anthropic's agentic coding tool — it reads files, runs commands, browses the web, and edits things on its own. Most SEO guides use one slice of that: they treat it as a smart drafting box. You ask, it writes, you paste the draft into WordPress yourself.

The difference in a real engine is that Claude Code owns every step in the chain, not just the middle one:

1. Pick the topic from a state file so it never repeats itself.

2. Research the competition — actually fetch the top-ranking pages and find their gaps.

3. Write the post against those gaps, in my voice, with real internal and external links.

4. Publish it to the live blog through the browser.

5. Measure and refill — read Search Console weekly, update the tracking sheet, top up the calendar.

Every guide I read covers step 3 well and hand-waves the rest — "then connect Zapier" or "then use our platform." Steps 1, 2, 4, and 5 are where the leverage is. They are also the boring 80% nobody wants to build. That is exactly why building them is worth it.

How the engine picks a topic without repeating itself

The single hardest problem in daily publishing is not writing. It is not writing the same thing twice. A model with no memory will happily give you six posts about "AI marketing tips" in a month.

I solve this with a plain markdown state file — a topics log. Every published post appends one row: date, title, slug, target keyword, and the source it came from. Before the engine writes anything, it reads that log. Its instructions are explicit: if a topic or angle already appears, derive a genuinely different sub-angle or move on. The log is also where it finds internal-link targets, so the blog builds its own link graph over time instead of orphaning every post.

This is the unglamorous trick behind "runs while I sleep." The intelligence is in the model; the reliability is in a text file that remembers what the model forgets.

How Claude Code researches competitors before it writes a word

This is the step that separates a post that ranks from a post that just exists. The engine does not start writing from its own knowledge. It starts by finding out what it is up against.

For a target keyword, it web-searches, pulls the three-to-six pages currently ranking, and fetches each one. Then it writes a short internal analysis: word count, which subtopics they cover, what data they cite, what format they use, and — the part that matters — where they are thin, outdated, or vague. When the job is heavy, it fans this out to subagents so several pages get read in parallel without clogging the main context.

Here is a real example from the post you are reading. I had the engine analyze the top guides for "claude code seo automation." Two of them were genuinely good — one ran about 8,500 words. But both had the same tell: when the topic got to actually running this daily and publishing automatically, they pivoted to "use our platform." The gap was the entire back half of the pipeline. That gap is why this post exists and what it is built to beat. You cannot find a gap like that from memory. You find it by reading the competition, which is what the engine does every single run.

How to publish automated blog posts with no API

Here is the part almost no guide covers, because it is the annoying part. Most CMS platforms — Podia, BeeHiiv, Wix, and plenty of others — do not give you a clean publishing API. The standard advice is to bolt on Zapier, a webhook, or a headless CMS. That is more moving parts to maintain and more monthly fees.

I skip all of it by publishing through the browser, the same way a person would. Claude Code drives a real Chrome session: opens the editor, types the title, pastes the formatted body, sets the slug, fills the excerpt, categories, and SEO fields, then clicks publish. There is no API key because there is no API. The blog you are reading this on was published exactly this way, by the engine, with me nowhere near the keyboard.

One concrete gotcha, since builder-shows-work means showing the ugly parts: getting formatted body text with live links into a rich-text editor is genuinely fiddly. Pasting raw markdown does not convert. The reliable path I landed on is to render the markdown to rich text first, copy it, and paste that — headings, bold, lists, and hyperlinks all land in one shot. It took a few failed runs to find that. That is the honest cost of "no API": you trade a monthly integration fee for a handful of browser quirks you solve once.

If you have read the daily marketing system I run every morning, this is the same philosophy pointed at SEO instead of social: let the agent operate the tools you already use, rather than forcing every tool to expose an API.

Does automated content get penalized by Google?

This is the fear that stops most people, so it is worth answering directly. The short version: Google does not penalize automation. It penalizes unhelpful content, whether a human or a machine made it.

Google's own guidance on creating helpful content is explicit. It asks you to create content "primarily for people," to be transparent about how it is made, and it flags "using automation extensively across many topics" and "summarizing others' work without adding value" as red flags. Notice what is not on that list: using AI. Google's stance is that how content is produced does not matter; whether it is genuinely useful and trustworthy does.

That is precisely why the competitor-research step is not optional. An engine that just spins out generic posts is exactly the low-value automation Google warns about. An engine that reads the top results, finds where they are weak, and publishes something demonstrably more useful and specific is doing what a good writer does — just without getting tired. The automation is the delivery mechanism. The value still has to be real, and the research step is what makes it real.

I keep a member of the loop human on purpose: I disclose that these posts are AI-assisted, and a few steps — like emailing the subscriber list — are hard-gated so the engine can never fire them on its own. Automation without a kill switch is how you wake up to a mess.

How the engine closes the loop: rank tracking and self-refilling calendars

Publishing every day is only half a system. The other half is knowing whether any of it worked and deciding what to write next.

Once a week the engine pulls real data from Google Search Console for each property — what is gaining impressions, what is stuck on page two, which queries are showing up that I never targeted. It writes that into a tracking file. Those page-two-and-climbing queries become next week's topics, so the content calendar refills itself from what is actually gaining traction instead of from a guess I made months ago. This measurement-to-calendar feedback loop is the piece every competitor guide leaves out entirely, and it is the difference between a content firehose and a system that gets smarter each week.

None of this needs me to be at my desk. Claude Code can run on a schedule — Anthropic-managed Routines keep working even when your computer is off, or a desktop scheduled task runs against your local files. Mine runs every weekday morning. I read the report over coffee.

Start small: one blog, one post, one week

You do not need my full three-blog setup to get value from this. The path that works:

• Week one: get Claude Code to research and draft one post for one blog. Just the writing, done well, with real competitor research. Publish it by hand.

• Week two: add the topics log so it stops repeating itself and starts internal-linking.

• Week three: automate the publishing step for your specific CMS — this is where browser publishing earns its keep.

• Week four: put it on a schedule and add the weekly rank review.

Each layer works on its own. You are never stuck with a half-built machine that does nothing until it is finished.

If you want the whole thing wired up for you — the routines, the copy-paste prompts, and the browser-publishing setup for Podia, BeeHiiv, and Wix — I walk through the entire engine in My Claude Code SEO Blog Engine live training. The replay is included, so you can build alongside it at your own pace.

The honest limits

A few things this engine does not do, because pretending otherwise helps no one. It does not invent original research or interview sources — it synthesizes what is already public, better. It will occasionally paste a body with a formatting quirk that needs a one-click fix. And it is only as good as the voice file and the competitor research behind it; point it at a thin keyword and you get a thin post. The engine removes the labor, not the judgment. You still decide what is worth publishing.

But for the actual grind — the daily research, drafting, formatting, and publishing that used to eat my mornings — it genuinely runs itself. This post is the proof. A machine picked the topic, read the competition, wrote the draft, and published it here while I did something else.

Build the whole revenue system, not just the blog

The SEO engine is one piece of a larger stack — the same one that markets my ~14 products. Inside the Reinventing AI Accelerator you get the live trainings that build these systems step by step, plus the growing vault of AI apps, agents, and tools members can deploy or sell. If you are trying to build and sell AI solutions without a team behind you, that vault is the shortcut. One small ask: come build one working thing, not ten half-finished ones.

FAQ

Can Claude Code actually publish a blog post on its own?

Yes — with the right setup. Claude Code can drive a real browser session, so it opens your CMS editor, enters the title and body, sets the slug and SEO fields, and clicks publish, no API required. In my engine it publishes to a live blog every weekday. I keep the final send-to-email-list step human-gated on purpose, but the publish itself is fully automated.

Do I need to know how to code to run an SEO engine like this?

You need to be comfortable giving clear instructions and running a terminal command, not to write software. The engine is built from markdown instruction files and prompts, not from code you maintain. If you can write a detailed brief for a freelancer, you can write the instructions Claude Code follows.

Will Google penalize AI-written blog posts?

No. Google's guidance targets unhelpful, low-value content regardless of how it was made, not AI use itself. The risk is publishing generic filler at scale. An engine that researches the current top results and publishes something more useful and specific is creating the people-first content Google rewards. Be transparent that it is AI-assisted, and keep the value real.

How is this different from an AI auto-blogging tool or plugin?

Auto-blogging tools spin out generic posts from a keyword with no idea what they are competing against. The difference here is the competitor-research step and the feedback loop: this engine reads the pages currently ranking, targets their specific gaps, and refills its calendar from real Search Console data each week. It is a research-and-publish system, not a text spinner.

How many posts a week is realistic for one person?

I publish one competitor-researched post per weekday, per blog, across three blogs — with no writer on payroll. Start with one post a week on one blog until the quality is where you want it, then add cadence. The bottleneck is never the writing; it is your review standard, which is exactly where a solo operator's time should go.