goog-geo — Google AI Search Readiness Audit Skill

A Claude Code skill that audits any website URL for Google AI Search readiness — whether your content can be crawled, indexed, excerpted, and cited by AI-powered search systems including Google AI Overviews, ChatGPT, Perplexity, Gemini, and Copilot.

Built directly from Google's official AI optimization guide, Google's AI features documentation, and the Princeton GEO research (KDD 2024).

---

What Is GEO?

Traditional SEO gets you ranked. GEO gets you cited.

In traditional search, you need to appear on page 1. In AI search, a well-structured page can get cited in an AI-generated answer even if it ranks third — because AI systems select sources based on content quality, structure, and extractability, not just PageRank.

Dimension	Traditional SEO	Generative Engine Optimization
Goal	Rank on page 1	Get cited in AI-generated answers
Selection signal	Link graph + keywords	Content quality, structure, extractability
Rank required?	Yes	No — a #3 page can be cited over #1
Key signals	Backlinks, PageRank	Semantic HTML, schema, answer blocks, authority
Crawl access	Googlebot + index/snippet eligibility	Cross-platform AI crawlers tracked separately
Content format	Keyword-optimized prose	Self-contained answer blocks, tables, FAQs
Freshness	Helps rankings	Critical — undated content loses to dated

---

What This Skill Does

Runs a live browser audit of any URL and produces a 100-point scored GEO report with:

• A letter grade (A–F)
• Per-category scores across 5 dimensions
• A pass/fail checklist of 28 specific checks
• An inspectable evidence bundle for technical checks: final URL, rendered title, crawler rule match, parsed robots directives, JSON-LD summary, and exact action plan
• A prioritized action plan (quick wins → medium effort → longer term)
• An explicit list of things not to do (common GEO myths debunked by Google)

The 5 Scoring Categories

Category	Points	What It Checks
Google Search AI Eligibility	25	Googlebot access, HTTP 200, noindex/nosnippet signals, canonical — the three hard gates for Google AI features
Helpful Non-Commodity Content	25	First-hand expertise, intent satisfaction, depth beyond commodity, AI content quality — scored by auditor judgment
Content Organization & Extractability	20	Heading hierarchy, direct answer block, FAQ sections, internal links, query fan-out coverage
Technical Structure & Page Experience	15	<main>, <article>, title/description, alt text, ARIA, Open Graph
Entity & Enhancement Signals	15	JSON-LD schema, author, publication date, statistics, cross-platform AI bot access

Grade Scale

Score	Grade	Meaning
90–100	A	AI-ready; only minor polish needed
75–89	B	Good foundation; targeted fixes will yield results
55–74	C	Significant gaps; structured improvement effort required
35–54	D	Major issues blocking AI visibility
0–34	F	Not AI-optimized; foundational work needed

---

Installation

Requirements

• Claude.ai with Skills enabled, or Claude Code
• Code execution enabled in Claude
• For Claude Code installs: playwright-cli available via npx (installed automatically if missing)

Install in Claude Web UI

You do not need a paid Claude plan to install this skill. Claude Skills are available on Free, Pro, Max, Team, and Enterprise plans, but Code execution must be enabled in Claude.

1. Open the goog-geo GitHub repository.
2. Click Code → Download ZIP.
3. In Claude, go to Customize → Skills.
4. Click the + button, choose Create skill → Upload a skill, and upload the ZIP file you just downloaded from GitHub.
5. Toggle the skill on.

Your uploaded skill is private to your Claude account unless you are on a Team or Enterprise plan and choose to share it with your organization.

Install in Claude Code

# 1. Clone the repo into your skills directory
git clone https://github.com/vishalmdi/goog-geo ~/.agents/skills/goog-geo

# 2. Install dependencies (required for scripts/audit.mjs)
cd ~/.agents/skills/goog-geo && npm install

# 3. Register it with Claude Code
ln -s ../../.agents/skills/goog-geo ~/.claude/skills/goog-geo

Already have ~/.agents/skills/ set up? Just run steps 1 and 2 — the symlink structure in step 3 matches how all skills are registered with Claude Code.

First-Time Browser Setup

The skill uses a headless Chromium browser to inspect the live rendered DOM. Install it via npm:

cd ~/.agents/skills/goog-geo
npm run install-browser

Or manually:

npx playwright install chromium --with-deps

---

Usage

In any Claude Code session, type:

/goog-geo

Claude will ask for a URL (or you can pass it directly):

/goog-geo https://yoursite.com/your-page

Claude will then:

1. Fetch robots.txt and HTTP headers
2. Open the URL in a headless browser
3. Extract headings, JSON-LD schema, meta tags, semantic HTML, and content signals
4. Attach evidence bundles to the key technical checks
5. Score all 5 categories
6. Return a full GEO report with action plan

You can also run the deterministic collector directly:

npm run audit -- https://yoursite.com/your-page

The collector outputs JSON with raw signals plus a checks object. Each check includes status, evidence, and actionPlan, so the score can be reviewed instead of treated as a magic grade.

Example Output

## Google AI Search Readiness Audit: https://example.com
**Overall Score: 72/100** — Grade: C
Audited: 2026-05-16

### Google AI Search Eligibility: ELIGIBLE
- Googlebot: allowed
- Indexing: allowed
- Snippets: allowed (no blocking directives)

### Score Breakdown
| Category                              | Score  |
|---------------------------------------|-------:|
| Google Search AI Eligibility          | 25/25  |
| Helpful Non-Commodity Content         | 14/25  |
| Content Organization & Extractability | 14/20  |
| Technical Structure & Page Experience | 11/15  |
| Entity & Enhancement Signals          |  8/15  |

### ✅ Passing Checks
- Googlebot not blocked in robots.txt
- Page returns HTTP 200
- No noindex or nosnippet directives
- Single H1 present and matches page intent
...

### ❌ Failed Checks — Highest Impact First
| Issue | Category | Impact | Recommended Fix |
|-------|----------|:------:|----------------|
| Content lacks first-hand expertise signals | Helpful Content | HIGH | Add named author with credentials; include original examples or data |
| No JSON-LD schema detected | Entity Signals | MEDIUM | Add Article + FAQPage schema via JSON-LD |
...

### Quick Wins (≤ 30 min)
1. Add named author attribution with credentials
2. Add publication and last-updated dates
3. Add JSON-LD Article schema

---

What NOT to Do (GEO Myths)

Google's AI optimization guide explicitly debunks these. The skill flags them and will tell you not to do them:

Myth	Reality
Add a llms.txt file	Google states this has no effect on AI search
Add special AI-targeted schema	Standard schema.org types are what matter
Chunk content into small pieces	Write for humans, not AI ingestion pipelines
Rewrite content specifically for AI	"Success often requires no overt SEO at all"
Build inauthentic citations	Manipulation is detected and penalized
Obsess over long-tail keywords	AI uses semantic understanding, not keyword matching

---

How This Was Built

Grounded in Google's Official Guide

This skill was built from Google's official documentation as the primary source of truth. A key principle from the guide: the same foundational signals that make content rank well in traditional search also make it citable in AI search. There is no separate technical path.

Google's AI Overviews and AI Mode use the same crawl, indexing, and ranking systems as traditional Search — with RAG (Retrieval-Augmented Generation) applied on top. The eligibility gates are:

1. Googlebot access — the only bot that matters for Google AI features; blocking it blocks all Google AI citation
2. Indexability — no noindex in meta tags or X-Robots-Tag HTTP headers
3. Snippet eligibility — no nosnippet or max-snippet:0; AI Overviews work by excerpting content

Beyond eligibility, content selection skews toward:

4. Unique, people-first content — genuinely helpful, non-commodity; "success often requires no overt SEO at all"
5. Structured, extractable content — answer blocks, tables, FAQs that AI can pull as self-contained passages
6. Entity recognition — schema, author attribution, and dates that help AI systems understand what a page is, who wrote it, and when

The Princeton GEO Research

The content quality scoring (Category 4) is grounded in the Princeton GEO research paper (Aggarwal et al., KDD 2024), which studied visibility across Perplexity.ai and identified the highest-impact optimization methods:

Method	Visibility Boost
Cite sources	+40%
Add statistics	+37%
Add expert quotations	+30%
Authoritative tone	+25%
Improve clarity	+20%
Technical terminology	+18%
Keyword stuffing	−10%

The key finding: low-authority sites benefit most — adding citations and statistics to a low-ranking page can increase AI visibility by up to 115%.

Technical Implementation

The skill uses playwright-cli (the headless browser CLI bundled with Playwright) to inspect the live rendered DOM. This is intentional and important:

• curl and web_fetch strip <script> tags, making them unable to detect JSON-LD schema injected by CMS plugins (Yoast, RankMath, AIOSEO)
• playwright-cli loads the full page in a real browser, exposing the complete accessibility tree — which is also what browser-based AI agents see when navigating your site

For automated data collection, the skill also ships scripts/audit.mjs — a deterministic Node.js script using the Playwright Node API that collects all DOM signals in a single browser session and outputs JSON. Claude then interprets the JSON and applies the scoring rubric.

The collector includes inspectable evidence for the highest-risk checks:

• Googlebot access — audited path, matched robots.txt user-agent, matched rule, and allow/block result
• Rendered page state — requested URL, final browser URL, navigation status, and rendered title
• Index/snippet controls — parsed meta robots and X-Robots-Tag directives, including noindex, nosnippet, and max-snippet
• Canonical URL — rendered canonical and whether it matches the final browser URL
• JSON-LD — rendered JSON-LD count, validity, and schema types

Scoring Design

The 5-category / 100-point framework (25/25/20/15/15) was designed to:

• Weight the three hard gates highest (Category 1, 25 pts): Googlebot access, noindex, and nosnippet together can completely prevent Google AI citation; they are weighted accordingly
• Weight helpful content second (Category 2, 25 pts): Google explicitly identifies unique, people-first content as the highest-leverage factor; this category requires auditor judgment, not DOM extraction
• Treat schema as an enhancement, not a gate: Schema supports rich results and entity clarity but is not required for AI Overviews; framing it as a blocker would be factually incorrect
• Keep partial scoring at every check: most real-world pages are somewhere in the middle

---

Related Skills

If you're using Claude Code with the full skill ecosystem:

Skill	Use After This Audit
ai-seo	Build a content strategy around the audit findings
seo-audit	Complement with a traditional technical SEO audit
schema-markup	Implement the JSON-LD fixes from Category 5
competitor-alternatives	Build comparison pages (highest-cited content type in AI answers)
programmatic-seo	Build AI-optimized content at scale

---

File Structure

goog-geo/
├── SKILL.md                                    # Main skill — loaded by Claude Code
├── package.json                                # Playwright dependency declaration
├── scripts/
│   └── audit.mjs                               # Deterministic data-collection script (outputs JSON)
└── references/
    ├── scoring-rubric.md                       # Detailed per-check scoring criteria and edge cases
    └── google-ai-search-principles.md          # Official guide summary — truth source for scoring

---

References

• Google Search Central: AI Optimization Guide
• Princeton GEO Research — Aggarwal et al., KDD 2024
• Google E-E-A-T Guidelines
• Schema.org
• Google Rich Results Test

---

Changelog

Version	Date	Changes
2.1.0	2026-05-18	Added inspectable evidence bundles to the deterministic collector; improved robots.txt parsing with URL-specific rules, user-agent specificity, longest-match precedence, and matched-rule evidence; added final rendered URL, parsed robots directives, JSON-LD summaries, canonical matching, npm run audit, and unit tests for parsing logic
2.0.0	2026-05-16	Repositioned as Google AI Search readiness auditor; corrected factual error (Google-Extended does not gate AI Overviews — Googlebot + noindex/nosnippet do); new 25/25/20/15/15 scoring model reflecting actual Google AI eligibility hierarchy; Category 2 adds qualitative helpful-content judgment; Category 3 adds query fan-out coverage; new scripts/audit.mjs for deterministic data collection; new references/google-ai-search-principles.md as truth source; eligibility verdict added to report header (ELIGIBLE / AT RISK / BLOCKED / UNKNOWN); schema reframed as enhancement signal, not requirement
1.1.0	2026-05-16	Corrected Google Search AI readiness scoring around Googlebot, indexability, and snippets; added informational bot notes for Google-Extended, Gemini-Bot, Meta-ExternalAgent, Applebot-Extended, and cohere-ai; added robots.txt HTML-body false-positive handling, X-Robots-Tag checks, max-snippet scoring, Top 3 Highest-Impact Fixes, sitemap detection, internal link and trust-page flags, JSON output mode, and preflight fallback warning
1.0.0	2026-05-16	Initial release

---

License

MIT