How to Run a Free SEO Audit With Claude Code (and Turn It Into a Fix List That Actually Ships)

Most "free SEO audit" tools hand you a score out of 100 and a wall of warnings, half of which don't matter for your site. That's not an audit. An audit is a decision-making document: what is wrong, how much it's costing you in rankings or clicks, and what to fix first.

We break a real audit into three passes. The first is crawl coverage: which pages exist, which ones search engines can reach, and which ones are stranded. The second is on-page relevance: does each page clearly answer one search intent. The third is technical health: can the page be indexed, does it load quickly, and is the markup clean. Claude Code is a good fit because it can read your sitemap, fetch your live pages, parse your HTML or template files, and hold all three passes in one place while it reasons about them.

Do not start by crawling your entire site. Start with the pages you'd be upset to lose. For a store or service site that's usually your money pages, your top blog posts, and your category or hub pages. Open your analytics and your search console, list the URLs that already get impressions or conversions, and hand that list to Claude Code first.

In Claude Code, point it at your sitemap.xml and ask it to compare the sitemap against your priority list. The questions you want answered are simple: are all my important pages in the sitemap, are there pages in the sitemap that shouldn't be public, and are there important pages that nothing links to. A prompt as plain as "Read sitemap.xml and my list of priority URLs in priority-pages.txt, then tell me which priority pages are missing from the sitemap and which sitemap URLs return non-200 status" gets you a real coverage map.

Crawling in priority order matters because attention is the scarce resource. Fixing the title tag on a page that earns clicks beats fixing fifty pages nobody visits.

On-page SEO comes down to one question per page: if someone searches the term this page should rank for, does the page obviously answer it? You can have Claude Code fetch a page and check the parts that signal intent to a search engine and a reader at the same time.

Walk through each priority page and have Claude pull the title tag, meta description, the single H1, the heading outline, the first paragraph, and the internal links pointing into and out of the page. Then ask whether they all point at the same topic. A common failure is a title promising one thing while the H1 and body drift somewhere else, or a page targeting a keyword that never appears in any heading. Another is thin content: a page that exists for a query but answers it in two sentences.

Ask Claude to be specific and quote the actual text it found, so you're reviewing evidence, not a verdict. "Here's the title, here's the H1, they disagree, here's why" is something you can act on. A bare score is not.

Technical issues are dangerous because they're invisible until traffic quietly drops. The highest-value checks catch pages that can't be indexed at all. Have Claude Code read each priority page's response headers and HTML head for a noindex tag, a robots directive, or a canonical tag pointing at a different URL. A page that accidentally canonicalizes to your homepage will never rank on its own, and nothing on the page tells you that.

Next, check robots.txt for rules that block sections you actually want crawled, and confirm your canonical URLs are consistent — pick either trailing-slash or non-trailing-slash and stick with it, http versus https, www versus non-www. Then look at load behavior: oversized images, render-blocking scripts, and templates that ship far more than the page needs. Claude can read the HTML and flag the heavy assets even without a full performance lab.

If you publish structured data, have Claude validate that the JSON-LD on the page is well-formed and matches the visible content. Markup that claims a review or price the page doesn't show is a liability, not a win.

An audit that ends in a list of 80 problems gets ignored. The last step is to force every finding into a small grid: impact (how much it could move rankings or clicks) against effort (how long it takes to fix). Ask Claude Code to take all the findings from the three passes and sort them into four buckets.

Do the high-impact, low-effort items this week — a noindex tag left on a money page, a broken canonical, a title that doesn't match its page. Schedule the high-impact, high-effort items, like rewriting thin content or restructuring an internal-linking hub. Batch the low-impact, low-effort fixes for when you're already in the file. And consciously drop the low-impact, high-effort work; deciding not to do something is part of the audit.

Have Claude output this as a plain checklist with the page URL, the exact problem, the proposed fix, and the bucket. That document is the actual deliverable. Everything before it was just gathering evidence.

The strength of running this in Claude Code is that the workflow is yours to keep. Save your prompts and your priority-URL list, and you can re-run the same audit monthly, or after any big site change, and diff this month's fix list against last month's. An audit you can repeat catches regressions before they cost you traffic.

If you'd rather not assemble the prompts and the crawl logic each time, the SEO Audit Agent skill packages this exact three-pass workflow — crawl priorities, on-page checks, technical checks, and the impact-versus-effort fix list — into one repeatable Claude Code skill. Run the manual version first so you understand what each check is doing; reach for the agent when you want the same audit on a schedule without rebuilding it by hand.