The SEO skills our audit uses: 6 analysis steps and 3 public helpers

When you run our audit, an AI agent works through six structured analysis skills against your site's data. Each produces specific, data-grounded findings, not generic advice. Here's exactly what each skill does, what data it consumes, and why it's in the audit. Three public helper skills also cover conversion-safe AI-copy editing, traffic trend analysis, and claim verification.

These are the same skills we used to audit this very site. The runbook and methodology are free to download. The audit runs on your machine, produces the report there, and then compiles a separate Action Runbook after report QA.

1. /programmatic-seo

What it does: Identifies scalable page types your site should build. For B2B SaaS, this means integrations pages, comparison pages, persona-based landing pages, glossary terms, and template directories.

Data it needs: Your site structure and business context — no GSC required. This is a discovery skill that works on any site.

Why it matters: On a thin site (the most common SaaS SEO problem), the biggest gap is that the pages your buyers search for don't exist, not that existing pages are bad yet. This skill builds the playbook.

Example output: Five launch-ready pSEO clusters with URL structures, content templates, and a prioritized launch order. For our own audit, it identified integrations, comparisons, personas, glossary, and directory as the highest-ROI clusters.

2. /ai-seo

What it does: Runs a fixed, bounded prompt-and-citation benchmark across the AI-search surfaces that matter to the client. It records the exact prompts, platform, date, locale, persona, observed brand mentions, client citations, and cited domains, then separates on-site opportunities from third-party visibility opportunities.

Data it needs: Your business context and top commercial queries. GSC data helps but isn't required.

Why it matters: AI-generated answers can introduce buyers to brands and sources, but their outputs vary by prompt, platform, market, and date. A reproducible panel gives you an observed baseline without pretending there is one universal “AI rank.” Google does not require a special AI file beyond normal Search eligibility, so the skill does not score or promise visibility from llms.txt.

Example output: A dated 20–50 prompt panel showing where the brand appeared, whether the client's domain was cited, which third-party domains recurred, and which validated on-site or off-site actions deserve testing next.

3. /site-architecture

What it does: Audits your information architecture and internal linking. Spoke-to-hub back-linking patterns, footer structure, pillar/cluster organization, orphan page detection, and navigation hierarchy.

Data it needs: Your full page list (from Squirrel Scan) and content clusters from the /programmatic-seo output. Works without GSC.

Why it matters: Internal linking is one of the few ranking levers you control completely. Orphan pages — pages with no internal links pointing to them — are invisible to both users and crawlers. Bad IA means even good content never gets found.

Example output: A restructured navigation tree, spoke-to-hub linking plan (every blog post links back to its pillar page, every pillar page links to a conversion page), and a list of orphan pages with recommended parent links. Our audit found an orphaned /phases/drift page and identified that the blog section was entirely missing from the nav.

4. /schema-markup

What it does: Validates every schema type on your site against schema.org and Google's requirements. Maps page types to their correct schema types, recommends @graph patterns for multi-type pages, and catches validation errors before they hit Search Console.

Data it needs: Squirrel Scan output — specifically the structured data errors and the list of schema types found on each page. Works without GSC.

Why it matters: Broken schema is worse than no schema. Google may still extract partial data from invalid markup, but the results are unpredictable. Common failures include: wrong domain in @id fields (our site had this: every schema URL referenced our staging domain), missing required properties, and using types Google doesn't support.

Example output: Per-page schema mapping, a recommended @graph structure, and a list of validation failures with fixes. Our audit identified 6 missing schema types and a domain mismatch that had been live since deploy.

5. /competitor-alternatives

What it does: Builds a comparison and alternative-page framework against your confirmed competitors. Covers four formats: direct A-vs-B comparisons, alternative-to pages, category comparison tables, and review mining for user-perceived gaps.

Data it needs: Confirmed competitor names from the client, never guessed. A comparison page built on the wrong competitor is worse than no page. Works without GSC.

Why it matters: "X vs Y" and "X alternatives" are the highest-intent bottom-of-funnel queries in B2B SaaS. Someone searching "Ahrefs alternative" is actively switching. If you're not on that page, your competitor is.

Example output: A prioritized comparison matrix, a centralized competitor data file (features, pricing, positioning, gaps), and 4-format content templates for each confirmed competitor. Our audit compared us against claude-seo.md, positioned as the privacy-first, no-terminal alternative.

6. /seo-audit

What it does: The synthesis skill. Runs a systematic SaaS-specific SEO gap checklist: content type gaps (are you missing comparison pages? case studies? integrations directories?), INP (Interaction to Next Paint) risk assessment, crawl and indexing evidence, and checks for originality, accuracy, first-party value, and policy compliance in existing content.

Data it needs: At least 4 weeks of GSC performance data for meaningful query and page analysis. This is the one skill that gates on data maturity. Running it on a brand-new site produces weak output.

Why it matters: This is the skill that connects the dots. The previous five skills produce findings in their lanes. This one cross-references them against each other and against your actual traffic data, dropping recommendations that don't connect to a specific data point.

Example output: A comprehensive audit with 27 on-page issues, 18 keyword opportunities, 9 content gaps, and a prioritized action plan split into 10 quick wins and 11 strategic investments. Every recommendation maps to a specific finding from the data.

Three public playbook helpers

These helpers are available on GitHub and skills.sh, with their Python scripts and reference files included:

• Remove AI slop without removing CTAs and good marketing copy — humanize AI-written blogs and marketing content while preserving calls to action, product links, offers, positioning, and proof. In Phase 2, the agent reviews audit copies when kickoff confirms AI-written pages.
• trend-scan — analyze Google Search Console CSV exports for traffic direction, declining pages and queries, ranking losses, and concentration risk before synthesis. The agent investigates the cause of each flagged change.
• claim-check — flag stale statistics, pricing, versions, feature claims, and missing verification dates before report delivery. The agent checks primary sources and records the evidence.

npx skills add BigChrisCooke/seo-playbook-skills

Select Claude Code, Cursor, or Codex in the installer. These three helpers require Python 3.10+ and supplement the core analysis skills. Their scripts can also run directly from a terminal. See the methodology installation guide for requirements and where each helper fits.

Run order matters

The skills are designed to be run in sequence, not in parallel:

1. /programmatic-seo — goes first because later skills reference its content clusters
2. /ai-seo — needs commercial queries but not dependent on other skill outputs
3. /site-architecture — depends on the page list and clusters from skill 1
4. /schema-markup — depends on Squirrel Scan data from Phase 1
5. /competitor-alternatives — needs confirmed competitor names (human-supplied)
6. /seo-audit — runs last, cross-referencing all previous skill outputs against GSC data

Prerequisite: Squirrel Scan must be run in Phase 1 before any skills. Without crawl data, skill outputs are generic. They can't cite real URLs and scores.

What happens after the skills run

The AI doesn't just dump six files and call it done. Phase 3 is synthesis, followed by a separate action handoff:

• All six skill outputs are collected into a staging document
• Duplicate findings are deduplicated (multiple skills often surface the same issue from different angles)
• Every recommendation is cross-referenced against raw data — recommendations that don't connect to a specific data point are dropped
• Findings are assigned impact scores (1–5), effort estimates (t-shirt sizes), and owners
• A 90-day plan is written first — it forces real prioritization
• Then the full REPORT.md is assembled following the methodology template
• Only after report QA passes, a separate ACTION-RUNBOOK.md maps every recommendation to research, AI capability, risk, approval, verification, cadence, and 2/4/8-week measurement

The result isn't just a health score. It is a client-ready diagnosis plus a living operating document. Every proposed fix keeps its “why,” owner, approval level, implementation evidence, and a way to verify whether it worked.

You can download the runbook to see exactly how each skill is invoked and what prompts the AI uses, read the full methodology for the rationale behind every step, or install the optional Claude Code skill pack for /seo-audit orchestration.