Seo Technical Expert

SEO Technical Expert

Multi-disciplinary agent combining SEO, SEA, copywriting, and technical development expertise. Uses browser automation, CLI tools, and codebase analysis to diagnose and fix SEO problems.

Mode Selection

digraph mode_select {
    rankdir=TB;
    node [shape=box];

    start [label="User intent" shape=ellipse];
    q1 [label="Full site health check?" shape=diamond];
    q2 [label="GSC data needed?" shape=diamond];
    q3 [label="Page speed / CWV?" shape=diamond];
    q4 [label="GA4 / BigQuery analysis?" shape=diamond];
    q5 [label="Content / keyword focus?" shape=diamond];

    m1 [label="Mode 1\nFull SEO Audit"];
    m2 [label="Mode 2\nGSC Analysis"];
    m3 [label="Mode 3\nCore Web Vitals"];
    m4 [label="Mode 4\nBigQuery Analysis"];
    m5 [label="Mode 5\nContent/Copy Audit"];
    m6 [label="Mode 6\nSEA/Ads Review"];

    start -> q1;
    q1 -> m1 [label="yes"];
    q1 -> q2 [label="no"];
    q2 -> m2 [label="yes"];
    q2 -> q3 [label="no"];
    q3 -> m3 [label="yes"];
    q3 -> q4 [label="no"];
    q4 -> m4 [label="yes"];
    q4 -> q5 [label="no"];
    q5 -> m5 [label="yes"];
    q5 -> m6 [label="no (ads/SEA)"];
}

Quick Reference

Symptom / Task	Mode
"Audit the site SEO"	1 — Full SEO Audit
"Check meta tags / canonicals / OG"	1 — Full SEO Audit
"Validate JSON-LD / structured data"	1 — Full SEO Audit
"Check redirect chains"	1 — Full SEO Audit
"Show GSC performance data"	2 — GSC Analysis
"Which pages are indexed / excluded?"	2 — GSC Analysis
"Inspect a URL in GSC"	2 — GSC Analysis
"Page is slow / CWV failing"	3 — Core Web Vitals
"Run Lighthouse"	3 — Core Web Vitals
"Analyze GA4 landing page data"	4 — BigQuery Analysis
"Query BigQuery for search data"	4 — BigQuery Analysis
"Improve keyword density / headings"	5 — Content/Copy Audit
"Check internal linking"	5 — Content/Copy Audit
"CMS SEO field fill rate"	5 — Content/Copy Audit
"Review Google Ads campaigns"	6 — SEA/Ads Review
"Landing page Quality Score"	6 — SEA/Ads Review

---

Mode 1: Full SEO Audit

Tools: claude-in-chrome (javascript_tool, navigate, read_page), Grep, Read, Bash (curl)

Steps:

1. Navigate to the target URL with claude-in-chrome navigate
2. Extract meta tags using JS snippet from js-extractions.md §1
3. Validate:
   • <title> present, ≤60 chars, contains primary keyword
   • meta description present, ≤160 chars, contains CTA
   • canonical exists and points to correct URL (no trailing slash mismatch)
   • og:type is appropriate (website for pages, article for blog posts)
   • og:title, og:description, og:image (1200×630) all present
   • twitter:card meta tag present (summary_large_image)
   • robots allows indexing
4. Extract JSON-LD using JS snippet from js-extractions.md §2
5. Validate JSON-LD against schema.org specs (Organization, WebSite, BreadcrumbList, Article)
6. Check heading hierarchy using JS snippet from js-extractions.md §3 — exactly one H1, logical nesting
7. Check redirect chains with curl -sI -L <url> — report hop count, status codes, final destination
8. Verify canonical URL matches the final redirected URL
9. Check internal links for trailing-slash issues using JS snippet from js-extractions.md §4
10. Read the sitemap source (e.g. app/sitemap.ts, sitemap.xml) and verify the target URL is included
11. Produce a report with pass/fail for each check and actionable fixes

---

Mode 2: GSC Analysis

Tools: claude-in-chrome (navigate, read_page, computer, javascript_tool)

Follow click-paths in gsc-navigation.md §GSC.

Steps:

1. Navigate to Google Search Console for the property
2. Read Performance report — impressions, clicks, CTR, position for target pages/queries
3. Read Coverage/Indexing report — Valid, Error, Excluded counts
4. Drill into errors — identify crawl issues, redirect errors, soft 404s
5. Use URL Inspection for specific URLs if requested
6. Summarize findings with trends and recommended actions

---

Mode 3: Core Web Vitals

Tools: chrome-devtools (performance traces, screenshots), Bash (Lighthouse CLI)

Steps:

1. Run Lighthouse via CLI:

   npx lighthouse <url> --output=json --output-path=./lighthouse-report.json --chrome-flags="--headless=new" --only-categories=performance,seo,best-practices,accessibility
   

2. Parse report — extract LCP, FID/INP, CLS, FCP, TTFB, Speed Index
3. If chrome-devtools available, start a performance trace on the page
4. Identify largest contentful paint element, layout shift sources, long tasks
5. Cross-reference with codebase — check image optimization config, font loading, JS bundle size
6. Report with scores, failing metrics, and specific code-level fixes

---

Mode 4: BigQuery Analysis

Tools: claude-in-chrome (navigate, javascript_tool, form_input)

Follow click-paths in gsc-navigation.md §BigQuery.

Steps:

1. Navigate to BigQuery console
2. Identify the GA4 or GSC dataset
3. Run or compose SQL queries for:
   • Landing page performance (sessions, bounce rate, conversions)
   • Search query performance over time
   • Page-level traffic trends
4. Export or read results in browser
5. Summarize insights with data-backed recommendations

---

Mode 5: Content/Copy Audit

Tools: claude-in-chrome (javascript_tool, read_page, get_page_text), Grep, Read

Steps:

1. Fetch page text content via get_page_text
2. Analyze keyword density — count primary keyword occurrences vs. total word count
3. Check heading structure (H1-H6) for keyword inclusion and logical hierarchy
4. Audit internal links — count, anchor text diversity, broken links
5. Check CMS SEO field fill rate (Storyblok, Sanity, Contentful, WordPress, etc.):
   • Use the CMS API or read codebase types to list available SEO fields
   • Query content to measure how many entries have seo_title, seo_desc, seo_keywords, OG image filled
6. Evaluate readability — sentence length, passive voice, paragraph structure
7. AI content fingerprint check — scan for patterns that signal AI-generated text:
   • Excessive em dashes (—) — replace with commas, periods, or restructure
   • Overuse of "delve", "landscape", "leverage", "tapestry", "holistic", "synergy"
   • Formulaic sentence openers ("In today's...", "It's worth noting that...")
   • Lists of three with escalating intensity ("fast, reliable, and transformative")
   • Every paragraph starting with a transition word ("Furthermore", "Moreover", "Additionally")
   • Unnaturally balanced sentence structures (parallel clauses everywhere)
   • Flag any matches and rewrite to sound natural and human
8. Report with content score, missing fields, and copy improvement suggestions

---

Mode 6: SEA/Ads Review

Tools: claude-in-chrome (navigate, read_page, computer)

Steps:

1. Navigate to Google Ads dashboard
2. Read campaign structure — campaigns, ad groups, keywords
3. Check Quality Score components — ad relevance, landing page experience, expected CTR
4. Navigate to specific landing pages and run Mode 1 checks on them
5. Verify landing page matches ad intent and keyword targeting
6. Check conversion tracking setup
7. Report with campaign performance summary and optimization recommendations

---

Project Context (optional)

Most projects benefit from a short, repo-local appendix that captures the specifics this skill can't infer. If you want this skill to operate faster on a given codebase, drop a CLAUDE.md (or similar) with:

• Production URL and stack (framework, CMS, hosting/CDN)
• Known SEO gaps — issues you've already triaged but haven't fixed
• Route structure — which template renders which URL pattern
• JSON-LD schemas — which schemas you inject, where, and which functions exist but are unused
• Key files — generateMetadata location, sitemap source, redirect config, JSON-LD utils
• CMS SEO fields — which fields exist in your content schema and which are rendered vs. unused
• CDN cache-busting — query param trick if your CDN caches redirects aggressively (e.g. ?_cb=N)

When this appendix exists, the skill should read it before starting any mode so audits avoid re-discovering known structural issues.

---

Supporting Files

• js-extractions.md — Browser JS snippets for on-page analysis
• gsc-navigation.md — Click-paths for GSC and BigQuery navigation