8 SEO bugs Claude Code found in my programmatic-SEO site before launch

A redirect rule that could 404 a whole site, 529 thin tag pages, self-comparison pages and more: what an AI audit of a static pSEO directory found, and the rule for each.

SkillNest ·

SkillNest is a programmatic-SEO directory: a Python script turns a JSON file of ~600 GitHub repositories into ~1,100 static pages on Cloudflare Pages. Programmatic SEO is great until it isn't — one wrong template rule is multiplied across every page.

Before submitting the sitemap, I asked Claude Code to audit the generated build folder the way a technical SEO would. Here is what it found, and the rule each fix turned into.

1. A redirect rule that could 404 the whole site

The generator wrote /* /404.html 404 into _redirects. On Cloudflare Pages, redirects are evaluated before static assets, so a catch-all rule can shadow real pages.

Rule: never use a catch-all rule for 404s on a host that already serves 404.html automatically.

2. Hundreds of one-item tag pages

529 of 665 tag pages listed exactly one item — textbook thin content.

Rule: group pages need at least 3 items to be indexable. Below that they get noindex,follow and stay out of the sitemap.

3. Self-comparison pages

The “A vs B” generator paired featured items with themselves, producing pages like docs vs docs.

Rule: only compare items within the same category, never an item with itself, and cap the total.

4. The same broken install command on every page

Every skill page showed /plugin install <slug>, which does not work for most skills.

Rule: generate per-item content from the data — here a real git clone command, with a different command when the skill lives in a sub-folder of a larger repository.

5. lastmod was the build date

Every URL in the sitemap had the same <lastmod>: the time of the build. Search engines learn to ignore that.

Rule: use each record's real update date.

6. Junk in the dataset

The weekly scraper had pulled in generic “awesome” lists and even a repository of leaked proprietary source code.

Rule: filter at build time, not by hand, so the next scrape cannot bring the junk back.

7. Thin structured data

Rule: SoftwareApplication + FAQPage + BreadcrumbList on item pages and ItemList on “best of” pages — using only facts that are visible on the page.

8. The domain move

The site moved from a free pages.dev subdomain to skillnest.dev.

Rule: 301 redirects that preserve path and query string, the new canonical host everywhere (sitemap, robots.txt, og:image), and Change of Address in Google Search Console.

The takeaway

Point the audit at the generator, not the generated HTML. Fix the HTML by hand and the next build quietly undoes your work.

The checks above are packaged as Claude Code skills in the SEO & Programmatic-SEO Skills Pack ($9) — but everything in this post works without it. If you are building something similar, the best SEO skills guide is a good place to start.