How to use Claude for SEO audits, the full walkthrough

An SEO audit with Claude is a process you install, never a prompt you improvise. This walkthrough runs all three levels on real sites, ours included, and shows the actual output at every step. No mockups, and one genuinely humbling score.

Using Claude for SEO audits means installing an audit skill, pointing it at a URL, a domain or a repo, and reviewing a report with a locked format: score, prioritised findings, quick wins, and an honest list of what it could not check. The three levels below go from a 10-minute baseline to fixes staged in git.

The three levels of Claude SEO audit

Pick by what you have access to. A URL anyone can audit. A domain you care about gets the technical sweep. A repo you own gets the deepest treatment, because Claude Code can read every file and stage the fixes.

Level 1: score a URL in ten minutes

Install Website Score and hand it your homepage. It grades the five factors Cyrus Shepard's 400+ site analysis found separate winning sites from losing ones, with evidence for every verdict.

We ran it on this very site. Here is the genuine output:

Ten minutes, a dated baseline, and three specific actions. That is what Level 1 buys you.

Level 2: the technical audit of a live domain

Install Technical SEO Audit and give it a domain. It fetches robots.txt, the sitemap and key pages live, runs status checks, and returns findings in priority order.

When we ran it against hawkacademy.co, our own education site, it found two critical issues professionals had missed for weeks:

Level 3: audit the codebase itself

If you own the repo, Claude Code SEO is the deepest level. It reads every HTML file, checks canonicals, schema, links and metadata against your live domain, writes SEO_REPORT.md, and offers the mechanical fixes as a commit you review. It never pushes.

The real run on lawrencehitches.com scored 90/100 across 589 files and still surfaced a case study shipping another article's meta description, five titles past 60 characters, and five category pages missing from every sitemap. Even good sites carry findings. The full excerpt lives on the Claude Code SEO page.

How to read a Claude audit report

Every audit skill in the library returns the same shape, and the shape is the discipline:

• The score header is your dated baseline. It only means something at the second audit.
• Priorities are the product. Critical this week, high this month, medium on the roadmap. If everything is urgent, nothing is.
• Verify before you fix. Findings name files and lines so you can look. Spot-check three before trusting the rest.
• Read the not-checked list. It names what needs a full crawler. An audit that admits limits is one you can act on, and one that does not is marketing.

Then apply the delegation matrix: the audit is delegated, the fixes are reviewed, the priorities stay yours.

The cadence that makes audits compound

One audit is a snapshot. The system is the re-run: baseline now, re-score after every meaningful change, the full technical sweep quarterly, and repo audits before deploys if you have the repo. Equal periods, same skills, same locked formats, so every report compares cleanly with the last one. The setup guide gets the skills installed, and the stack guide covers the exports that sharpen everything.

FAQ

How long does a Claude SEO audit take?

Level 1 takes about ten minutes for a scored baseline. Level 2 runs in roughly half an hour against a live domain. Level 3 depends on repo size: small sites get every page read, large repos get a labelled representative sample. All three are faster than the meeting where you argue about what to fix first.

Can I trust the findings?

Verify them, which is the point of how the skills report. Every finding names the URL, file or line where it was observed, so you can look. Each report also ends with an honest not-checked list naming what needs a full crawler. Trust the audit because you spot-checked it, never because it sounded confident.

What do I need before I start?

A URL is enough for Levels 1 and 2, the skills fetch what they need live. Level 3 needs Claude Code running inside your website's git repository. Search Console exports are optional at every level but they sharpen the diagnosis, and the stack guide covers exactly what to export.

Is this really free?

The skills are MIT licensed and free forever, and Anthropic makes skills available on every Claude plan including Free with code execution enabled. The honest total cost is your review time, because no audit at any price should ship fixes to production without a human reading the diff.