GEO Audit Orchestration Skill

Purpose

This skill performs a comprehensive Generative Engine Optimization (GEO) audit of any website. GEO is the practice of optimizing web content so that AI systems (ChatGPT, Claude, Perplexity, Gemini, etc.) can discover, understand, cite, and recommend it. This audit measures how well a site performs across all GEO dimensions and produces an actionable improvement plan.

What this composite is, and what it isn't. This score is a readiness measure: it reads public signals (content, schema, crawler access, off-page authority) and infers how citable and recommendable the site is. It does not query the AI engines to confirm the business is actually named in their answers. Read it as "how well-built for AI is this site," not "is this site in the answer right now." For the live outcome, run the free scan at areyoufoundbyai.com: two buyer questions on ChatGPT and Gemini, no signup; Pro asks up to 25 buyer questions across seven engines on the schedule you choose, from a shared token balance, and its headline measure is the named share of saved answers (the two are complementary: readiness here, visibility there; see docs/THL-GEO-METHOD.md). This composite is the open audit's own formula. It is not the product's AI Readiness score, which is built differently (five site signals, with the off-site Footprint scored separately). Every dimension below carries a provenance tag; a [heuristic] tag means model judgement from signals, not a measured fact.

Key Insight

Traditional SEO optimizes for search engine rankings. GEO optimizes for AI citation and recommendation. Sites that score high on GEO metrics see 30-115% more visibility in AI-generated responses (Georgia Tech / Princeton / IIT Delhi 2024 study). The two disciplines overlap but have distinct requirements.

THL enhancements (this fork)

Tech Horizon Labs runs this audit as part of a three-layer method (see docs/THL-GEO-METHOD.md):

• External benchmark. Alongside the dimensional composite below, run the agent-readiness-scan skill (THL-original) for Cloudflare's independent isitagentready.com 0 to 100 score. Record both; on a re-audit, track the delta on each: the movement is the proof, not the first number.
• Checklist. Work through references/thl-audit-checklist.md so no dimension is silently skipped and the same facts/scores stay consistent across every section.
• Deliverable. Assemble the audit data and run tools/audit-report-kit (THL-original) to produce a branded client PDF + compile-checked JSON-LD for the schema fixes, instead of leaving a raw markdown file.

Audit Workflow

Phase 1: Discovery and Reconnaissance

Step 1: Fetch Homepage and Detect Business Type

1. Use WebFetch to retrieve the homepage at the provided URL.
2. Extract the following signals:
   • Page title, meta description, H1 heading
   • Navigation menu items (reveals site structure)
   • Footer content (reveals business info, location, legal pages)
   • Schema.org markup on homepage (Organization, LocalBusiness, etc.)
   • Pricing page link (SaaS indicator)
   • Product listing patterns (E-commerce indicator)
   • Blog/resource section (Publisher indicator)
   • Service pages (Agency indicator)
   • Address/phone/Google Maps embed (Local business indicator)
3. Classify the business type using these patterns:

Step 2: Crawl Sitemap and Internal Links

1. Attempt to fetch /sitemap.xml and /sitemap_index.xml.
2. If sitemap exists, extract up to 50 unique page URLs prioritized by:
   • Homepage (always include)
   • Top-level navigation pages
   • High-value pages (pricing, about, contact, key service/product pages)
   • Blog posts (sample 5-10 most recent)
   • Category/landing pages
3. If no sitemap exists, crawl internal links from the homepage:
   • Extract all <a href> links pointing to the same domain
   • Follow up to 2 levels deep
   • Prioritize pages linked from main navigation
4. Respect robots.txt directives -- do not fetch disallowed paths.
5. Enforce a maximum of 50 pages and a 30-second timeout per fetch.

Step 3: Collect Page-Level Data

For each page in the crawl set, record:

• URL, title, meta description, canonical URL
• H1-H6 heading structure
• Word count of main content
• Schema.org types present
• Internal/external link counts
• Images with/without alt text
• Open Graph and Twitter Card meta tags
• Response status code
• Whether the page has structured data

Phase 2: Parallel Subagent Delegation

Delegate analysis to 5 specialized subagents. Each subagent operates on the collected page data and produces a category score (0-100) plus findings.

Return contract (THL), every subagent follows it: return your dimension score(s) as a number 0-100, a provenance tag ([scan] = data fetched this run · [partial-scan] · [heuristic] = judgement, no data · [unmeasured] = source unavailable → emit n/a, not a number), and findings each tied to the evidence that justifies it. Return the orchestrator's named dimensions directly. Do not invent your own blended sub-composite or your own weights (that breaks the fixed composite formula below).

Subagent 1: AI Visibility Analysis (geo-ai-visibility)

• Return two separate dimension scores: AI Citability (quotability/extractability) and Brand Authority (entity recognition), not one blended "AI visibility" number, since the composite weights them differently (25% vs 20%).
• Analyze content blocks for quotability by AI systems (citability scoring)
• Check AI crawler access via robots.txt and llms.txt presence (these also inform Technical GEO)
• Scan brand presence across YouTube, Reddit, Wikipedia, LinkedIn for the Brand Authority score

Subagent 2: Platform Optimization (geo-platform-analysis)

• Assess readiness for Google AI Overviews, ChatGPT, Perplexity, Gemini, Bing Copilot
• Check platform-specific ranking factors and optimization opportunities

Subagent 3: Technical GEO Infrastructure (geo-technical)

• Analyze robots.txt for AI crawler access
• Verify meta tags, headers, and technical accessibility for AI systems
• Check page speed, server-side rendering, and Core Web Vitals
• Assess security headers and mobile optimization

Subagent 4: Content E-E-A-T Quality (geo-content)

• Evaluate Experience, Expertise, Authoritativeness, Trustworthiness signals
• Check author bios, credentials, source citations
• Assess content freshness, depth, and originality
• Verify "About" page quality and team credentials

Subagent 5: Schema & Structured Data (geo-schema)

• Validate all schema.org markup
• Check for GEO-critical schema types (FAQ, HowTo, Organization, Product, Article)
• Assess schema completeness and accuracy
• Identify missing schema opportunities

Phase 3: Score Aggregation and Report Generation

Composite GEO Score Calculation

The overall GEO Score (0-100) is a weighted average of six category scores:

Formula:

GEO_Score = (Citability * 0.25) + (Brand * 0.20) + (EEAT * 0.20) + (Technical * 0.15) + (Schema * 0.10) + (Platform * 0.10)

Score Interpretation

Issue Severity Classification

Every issue found during the audit is classified by severity:

Critical (Fix Immediately)

• All AI crawlers blocked in robots.txt
• No indexable content (JavaScript-rendered only with no SSR)
• Domain-level noindex directive
• Site returns 5xx errors on key pages
• Complete absence of any structured data
• Brand not recognized as an entity by any AI system

High (Fix Within 1 Week)

• Key AI crawlers (GPTBot, ClaudeBot, PerplexityBot) blocked
• No llms.txt file present
• Zero question-answering content blocks on key pages
• Missing Organization or LocalBusiness schema
• No author attribution on content pages
• All content behind login/paywall with no preview

Medium (Fix Within 1 Month)

• Partial AI crawler blocking (some allowed, some blocked)
• llms.txt exists but is incomplete or malformed
• Content blocks average under 50 citability score
• Missing FAQ schema on pages with FAQ content
• Thin author bios without credentials
• No Wikipedia or Reddit brand presence

Low (Optimize When Possible)

• Minor schema validation errors
• Some images missing alt text
• Content freshness issues on non-critical pages
• Missing Open Graph tags
• Suboptimal heading hierarchy on some pages
• LinkedIn company page exists but is incomplete

Output Format

Generate a file called GEO-AUDIT-REPORT.md with the following structure:

# GEO Audit Report: [Site Name]

**Audit Date:** [Date]
**URL:** [URL]
**Business Type:** [Detected Type]
**Pages Analyzed:** [Count]

---

## Executive Summary

**Overall GEO Score: [X]/100 ([Rating])**

[2-3 sentence summary of the site's GEO health, biggest strengths, and most critical gaps.]

### Score Breakdown

| Category | Score | Weight | Weighted Score | Provenance |
|---|---|---|---|---|
| AI Citability | [X]/100 | 25% | [X] | [scan] |
| Brand Authority | [X]/100 | 20% | [X] | [scan] |
| Content E-E-A-T | [X]/100 | 20% | [X] | [partial-scan] |
| Technical GEO | [X]/100 | 15% | [X] | [scan] |
| Schema & Structured Data | [X]/100 | 10% | [X] | [scan] |
| Platform Optimization | [X]/100 | 10% | [X] | [scan] |
| **Overall GEO Score** | | | **[X]/100** | |

**Provenance vocabulary (THL enhancement, every row carries one):** `[scan]` = scored
from data actually fetched this run · `[partial-scan]` = some pages sampled, others
inferred · `[heuristic]` = model judgement, no underlying data fetched · `[unmeasured]`
= the data source needed was unavailable. When a category is `[unmeasured]` or pure
`[heuristic]`, emit `n/a` for its score, **never a number**: a number with weak
provenance still reads as hard data. This is how the audit stays honest about what it
actually measured.

---

## Critical Issues (Fix Immediately)

[List each critical issue with specific page URLs and recommended fix]

## High Priority Issues

[List each high-priority issue with details]

## Medium Priority Issues

[List each medium-priority issue]

## Low Priority Issues

[List each low-priority issue]

---

## Category Deep Dives

### AI Citability ([X]/100)
[Detailed findings, examples of good/bad passages, rewrite suggestions]

### Brand Authority ([X]/100)
[Platform presence map, mention volume, sentiment]

### Content E-E-A-T ([X]/100)
[Author quality, source citations, freshness, depth]

### Technical GEO ([X]/100)
[Crawler access, llms.txt, rendering, headers]

### Schema & Structured Data ([X]/100)
[Schema types found, validation results, missing opportunities]

### Platform Optimization ([X]/100)
[Presence on YouTube, Reddit, Wikipedia, etc.]

---

## Quick Wins (Implement This Week)

1. [Specific, actionable quick win with expected impact]
2. [Another quick win]
3. [Another quick win]
4. [Another quick win]
5. [Another quick win]

## 30-Day Action Plan

### Week 1: [Theme]
- [ ] Action item 1
- [ ] Action item 2

### Week 2: [Theme]
- [ ] Action item 1
- [ ] Action item 2

### Week 3: [Theme]
- [ ] Action item 1
- [ ] Action item 2

### Week 4: [Theme]
- [ ] Action item 1
- [ ] Action item 2

---

## Appendix: Pages Analyzed

| URL | Title | GEO Issues |
|---|---|---|
| [url] | [title] | [issue count] |

Quality Gates

• Page Limit: Never crawl more than 50 pages per audit. Prioritize high-value pages.
• Timeout: 30-second maximum per page fetch. Skip pages that exceed this.
• Robots.txt: Always check and respect robots.txt before crawling. Note any AI-specific directives.
• Rate Limiting: Wait at least 1 second between page fetches to avoid overloading the server.
• Error Handling: Log failed fetches but continue the audit. Report fetch failures in the appendix.
• Content Type: Only analyze HTML pages. Skip PDFs, images, and other binary content.
• Deduplication: Canonicalize URLs before crawling. Skip duplicate content (e.g., HTTP vs HTTPS, www vs non-www, trailing slashes).

Business-Type-Specific Audit Adjustments

SaaS Sites

• Extra weight on: Feature comparison tables (high citability), integration pages, documentation quality
• Check for: API documentation structure, changelog pages, knowledge base organization
• Key schema: SoftwareApplication, FAQPage, HowTo

Local Businesses

• Extra weight on: NAP consistency, Google Business Profile signals, local schema
• Check for: Service area pages, location-specific content, review markup
• Key schema: LocalBusiness, GeoCoordinates, OpeningHoursSpecification

E-commerce Sites

• Extra weight on: Product descriptions (citability), comparison content, buying guides
• Check for: Product schema completeness, review aggregation, FAQ sections on product pages
• Key schema: Product, AggregateRating, Offer, BreadcrumbList

Publishers

• Extra weight on: Article quality, author credentials, source citation practices
• Check for: Article schema, author pages, publication date freshness, original research
• Key schema: Article, NewsArticle, Person (author), ClaimReview

Agency/Services

• Extra weight on: Case studies (citability), expertise demonstration, thought leadership
• Check for: Portfolio schema, team credentials, industry-specific expertise signals
• Key schema: Organization, Service, Person (team), Review

Professional / financial services (YMYL)

• Extra weight on: displayed credentials (CFP/FChFP/degrees), regulatory IDs (AFSL / AR number / registration), named article authors, and trust/compliance pages (FSG, privacy, complaints).
• Check for: each practitioner has on-page credentials and a Person node; the firm uses the right LocalBusiness subtype (not bare Organization); YMYL claims are sourced, not asserted.
• Key schema: FinancialService / ProfessionalService / LegalService / MedicalBusiness (the correct LocalBusiness subtype), Person (per adviser), FAQPage.
• Entity reconciliation: identify the legal entity vs the trading/brand/domain name (they often differ) so the audit findings and the schema name one consistent entity.