How to use Claude for SEO, step by step

Claude does real SEO work when you stop treating it like a chatbot and start treating it like a junior with a checklist. Here is the exact setup we run at StudioHawk, from install to output, using the free Claude SEO Skills library.

Using Claude for SEO means three things: installing skill files that encode repeatable workflows, feeding Claude your real data (Search Console exports, crawl files, live URLs), and reviewing its findings before anything ships. Set up that way, Claude runs audits, briefs, clustering and reporting at agency pace. This guide covers the whole loop.

What Claude can actually do for SEO

The honest capability list, from running it across 500+ client campaigns:

• Audits. Technical audits from live fetches, codebase audits from a repo, store audits from three URLs. Findings with file paths and line numbers, prioritised.
• Data analysis. Hand it a Search Console export and it finds cannibalisation, decay and update damage that would take you an afternoon in spreadsheets.
• Content planning. Briefs, intent mapping, keyword clustering, topical maps. The planning layer before anyone writes.
• Schema and code. Valid JSON-LD generated and validated, and with Claude Code, fixes staged as commits you review.

What it cannot do: track rankings live, crawl the whole web, or know yesterday's algorithm update. You bring the data. It brings the process.

Step 1: install skills instead of hoarding prompts

A skill is a markdown file that teaches Claude a workflow: when to trigger, which steps to run, what to output. Install once, and the tenth audit follows the same process as the first.

git clone https://github.com/lhitches/claude-seo-skills.git
mkdir -p ~/.claude/skills/
cp claude-seo-skills/skills/*.md ~/.claude/skills/
cp -r claude-seo-skills/skills/topical-authority-map ~/.claude/skills/

That installs all 39. Or pull one skill with curl from any page in the directory. No install at all also works: paste a skill file into a chat and the process runs manually.

Step 2: feed it your real data

Claude with no data gives you generic advice. Claude with your data gives you findings. The three inputs that change everything:

• Search Console exports. Queries with the page dimension added. This switches on cannibalisation detection, decay diagnosis and update triage.
• Crawl data. A Screaming Frog export, or let the skill fetch your robots.txt, sitemap and key pages live. Full source-to-skill mapping in our stack guide.
• Your own site's repo if you have one. Claude Code SEO reads every HTML file and writes SEO_REPORT.md with staged fixes.

Step 3: run the first three jobs

Job one: score your site. Run Website Score on your homepage for a baseline. Ten minutes, and you know where you stand.

Job two: the real audit. Point Technical SEO Audit at your domain. When we ran it on our own education site, it found the main commercial page canonicalling to an A/B test variant and a dead www subdomain returning errors. Two critical issues on a site professionals maintain. Yours has something too.

Job three: plan content. Feed Information Gain a draft and its target query, and it shows you line by line where you are repeating the consensus and what only you can add.

What to keep human

Strategy, prioritisation trade-offs, and anything that ships to production. Claude outputs findings for review, never unsupervised changes. The full delegation matrix is its own guide: which SEO tasks to delegate to Claude.

The mistakes that waste your first week

• Asking without data. "How do I improve my SEO" gets you a blog post. An export gets you a diagnosis.
• Re-explaining the process every session. That is what skills exist to fix. Install them once.
• Shipping without review. Claude is confident when right and confident when wrong. The review step is not optional, it is the job.
• Treating one good output as a system. One great audit proves capability. A skill makes it repeatable. Repeatable is what compounds.

FAQ

Is Claude actually good at SEO?

Yes, for process work. Audits, briefs, clustering, schema and reporting all follow checkable steps, and Claude runs steps well. It has no live rank tracker and no crawl index of its own, so you bring the data and it brings the analysis. That trade works strongly in your favour.

Do I need Claude Code or is the web app enough?

The web app and desktop app run every skill in the library. Claude Code adds one thing the others cannot do: it reads an entire website codebase from the CLI and stages fixes as git commits. If you manage your own site's repo, that one is worth the setup.

How much does this cost to run?

The skills are free and MIT licensed. Anthropic makes skills available on every Claude plan including Free, with code execution enabled. A paid plan buys you longer sessions and more capacity, which matters once you are feeding it full crawl exports.

Where do I start if I only have 20 minutes?

Install one skill and run it on your own site. Website Score takes a URL and gives you a scored baseline in minutes. The first report usually names two or three things you did not know were broken, and now you have a to-do list that justifies the next 20 minutes.