30 Days of Using Claude Code to Grow a Real SEO Site

This site runs its own SEO and content work through Claude Code — not as a demo, as the actual day-to-day process. Thirty days in, here's what that honestly looked like: what got built, what broke along the way, and what — despite all of it — still hasn't converted a single real visitor.

1 What Actually Got Built

Thirty Claude Code guides, organized into one pillar page instead of left as scattered posts. Five new free tools alongside an existing one, each with a working implementation — the contrast checker's WCAG math was checked against known reference values before shipping, not assumed correct. Every batch of content edits ran through a local SEO audit script before deploying, checking titles, meta description length, duplicate content, and broken internal links.

2 What Broke — and How It Got Caught

A batch content edit corrupted 9 pages. A regex meant to replace one block of content matched too little and left old and new content concatenated together instead of cleanly swapped. It wasn't caught by the edit itself — it was caught by a verification pass afterward that checked for leftover text that should have been gone.

The actual lesson isn't "regex is risky," it's that automated edits at scale need automated verification at the same scale — checking after every batch, not just trusting that an edit that ran without an error also did the right thing.

3 The Honest Number: Zero Real Conversions So Far

The free guide's signup flow was reviewed against real delivery logs this sprint — every signup this month came from the site's own test addresses, not a real third-party visitor. That's not a failure of the tracking (the event fires correctly, verified against the actual delivery records) — it's just where a new domain with limited backlinks and no paid traffic honestly sits 30 days in. Content and tooling compound slowly; this post isn't claiming otherwise.

4 What Turned Out to Need a Human Anyway

Outreach emails, Dev.to publishing, a Product Hunt launch, GitHub pull requests — all of it got fully researched and drafted: real dead links found and verified via direct HTTP checks, real target repos confirmed active, exact copy written and ready to paste. None of it could actually be sent, because every one of those steps needs a real account and a real identity behind it. Claude Code can prepare the work completely; it can't be the person hitting send on someone else's platform.

Frequently Asked Questions

Did traffic actually grow in 30 days?

That's not in this post, on purpose — a new domain with limited backlinks doesn't show a meaningful ranking or traffic signal in 30 days regardless of how much content goes out. What this post covers is the work itself: what got built, what broke, and what's honestly still at zero.

What was the biggest mistake?

A regex bug during a batch content edit that concatenated old and new content on 9 pages instead of cleanly replacing it — caught by a post-edit verification check, not by the edit process itself, which is the actual lesson: verify after every batch, don't trust that an edit did what it looks like it did.

What tasks turned out to be impossible to fully automate?

Anything requiring a real account on someone else's platform — sending outreach email, publishing to Dev.to, submitting a Product Hunt launch, opening a GitHub PR. All of that got fully researched and drafted, then handed off for a human to actually send, because that step needs an identity a script can't have.

Is this a sales pitch disguised as a case study?

It's an honest account, which includes the parts that don't flatter the pitch — zero real third-party leads so far, several bugs that shipped before being caught, and a list of tasks that turned out to need a human either way. A case study that only shows wins isn't a case study.