seo-geo-skill

_{Real output. Reproduce it: python scripts/benchmark_demo.py}

Run it on your site in 10 seconds

curl -fsSL https://raw.githubusercontent.com/yanivgoldenberg/seo-geo-skill/main/seo-geo.md \
  -o ~/.claude/skills/seo-geo.md

# then in Claude Code:
/seo-geo https://yoursite.com

One command. Full audit. Ranked fix list. Works on any CMS.

We benchmarked 61 top SaaS and AI sites. 67% fail AI Search Readiness.

Two AI category leaders, OpenAI and Perplexity, score 7/100 each. Railway scores 0. Full leaderboard in State of AI Search Visibility 2026.

The bottom of the 61-site leaderboard tells the story:

The AI category leaders score near zero on AI Search Readiness.

Full 61-site leaderboard + methodology | OpenAI case study: 7 to 70 fix blueprint

What the data shows

• 67% of 61 top SaaS and AI sites scored 60 or below. Failing AI Search Readiness is the norm.
• Mean score: 49.4 / 100. Most sites leave the majority of rubric points on the table.
• llms.txt is entity hygiene, not a guaranteed ranking lever - Google has stated no special file is required for AI Overviews. But the sites that publish it consistently pair with the strongest GEO scores.
• 73% have no proper Person or Organization schema on their homepage.
• OpenAI and Perplexity score 7/100 each. The category leaders are almost invisible to their own category of search.

If your site scores below 60 on this rubric, you're in the bottom third of well-funded SaaS brands. The three cheapest wins:

1. Publish /llms.txt and /llms-full.txt → +25 points
2. Add Organization + Person JSON-LD → +15 points
3. Allow AI crawlers explicitly in robots.txt → +10 points

This skill automates all three in one session.

Original 13-site cohort (the first public benchmark)

Before the 61-site study, I scored a 13-site cohort and published it in docs/benchmarks.md. Different run, different rubric calibration (v1.6.0), so the scores here are not the same numbers as the 61-site study above. This is the original Top-13, reproduced here so the two documents always agree:

Full per-dimension breakdown and methodology in docs/benchmarks.md.

What Google actually says (and what this score is)

I am not going to sell you a guarantee. Google has publicly stated that no special file is required to appear in AI Overviews, and there is no documented ranking signal called "llms.txt." So this is the honest framing:

• The score is AI Search Readiness: machine-legibility, not a ranking promise. It measures how cleanly a machine (a crawler, an LLM, an answer engine) can parse your entity, your schema, and your stated facts. It does not predict your position.
• Citations and rankings are the outcome, not the deliverable. I make your site easier to read and trust. Whether ChatGPT, Claude, Perplexity, or Google AI Overviews then cite you depends on their systems, your authority, and your content. I do not control those, and anyone who says they guarantee a #1 result is selling you something I will not.
• The evidence is correlation, stated as correlation. Across the 61-site benchmark, the sites that publish /llms.txt consistently pair with the strongest GEO scores. That is a pattern in the data, not a causal lever I am promising.

What I will stand behind: the score is reproducible. Run tests/benchmark_sites.py against any URL and you get the same number I do. No claim in this repo lacks a number or a named site behind it.

Who this is for

• Marketers who want citations in ChatGPT, Perplexity, and Google AI Overviews
• Developers adding SEO to a client site and need it done right the first time
• Agencies running the same 20-phase audit on 50+ client sites per year

Not a fit if you want a 20-sub-skill toolbox. See docs/compare.md for honest comparisons to other Claude Code SEO skills.

What it does (Phase 0 audit + 20 optimization phases)

Open seo-geo.md for the full skill. At a glance:

The audit (Phase 0, non-destructive, always safe):

• 100-point rubric across 6 dimensions: Technical 20 + On-Page 15 + Schema 20 + GEO 25 + AEO 10 + E-E-A-T 10
• Same rubric is used by tests/benchmark_sites.py; no methodology drift between docs and code
• Phases 1-3 - Technical SEO, on-page, schema (16 types)
• Phase 4 - LLM citation (llms.txt + entity anchoring)
• Phase 5 - Answer engine optimization
• Phase 6 - E-E-A-T trust signals
• Phases 7-14 - Content, speed, hreflang, debugging

The writes (opt-in via --apply):

• Phase 15 - WordPress security hardening
• Phase 16 - Field patterns from production
• Phase 17 - Dry-run safety gates + banned endpoint list
• Phase 18 - Multi-platform adapters (WordPress / Shopify / Webflow / Next.js)
• Phase 19 - Competitor benchmarking
• Phase 20 - LLM Extractability Polish (LLM Summary, claims table, cite block, glossary, raw-data download, self-consistency check)

Install

# Global install
curl -fsSL https://raw.githubusercontent.com/yanivgoldenberg/seo-geo-skill/main/seo-geo.md \
  -o ~/.claude/skills/seo-geo.md

# Project-level only
curl -fsSL https://raw.githubusercontent.com/yanivgoldenberg/seo-geo-skill/main/seo-geo.md \
  -o .claude/skills/seo-geo.md

Usage

# Start here
/seo-geo https://yoursite.com

# Variants
/seo-geo --verify                # self-test: tools accessible?
/seo-geo --audit-only            # score only, no writes
/seo-geo --phase geo             # LLM citation only (fastest ROI)
/seo-geo --phase technical       # technical SEO only
/seo-geo --page <url>            # single page audit
/seo-geo --apply                 # writes enabled (opt-in, see Phase 17)
/seo-geo --benchmark <competitor> # head-to-head score

Seen at / Running in production

• yanivgoldenberg.com - canonical reference deployment. 94/100 single-page, 97/100 with the benchmark's 4-page sampling, both on live 2026-06-05 runs, rank 1 of 61 on the same rubric. The monthly Action commits every fresh snapshot to docs/state-of-ai-search-history/. Run it yourself and you get today's number. (proof)
• Used to generate the State of AI Search Visibility 2026 benchmark (61 SaaS and AI sites, 67% fail).
• Public benchmark methodology: tests/benchmark_sites.py - runnable and reproducible by anyone.

Schemas supported

Person Organization SoftwareApplication LocalBusiness Product Service Article BlogPosting FAQPage HowTo BreadcrumbList WebSite ProfilePage DefinedTerm Review SpeakableSpecification VideoObject

Author

Yaniv Goldenberg | LinkedIn

Fractional Head of Growth. Scaled a PLG SaaS to 20M+ users (100x ARR), tripled MRR at a consumer SaaS (+337%), grew an MLOps SaaS acquired by a Fortune 100. $100M+ in ad spend managed. I help post-PMF SaaS and e-commerce brands dominate AI search results - so ChatGPT, Claude, and Perplexity cite you before your competitors.

If this saved you an afternoon, star the repo.

Want this applied to your SaaS site?

I run a paid AI Search Visibility Audit for post-PMF SaaS, B2B, and e-commerce brands.

For post-PMF SaaS and e-commerce brands with organic demand but no technical citation layer. Not keyword research. Not link building. The layer that helps AI systems understand, trust, and cite your brand.

You get:

• 0-100 AI Search Readiness score across 6 dimensions (Technical 20 + On-Page 15 + Schema 20 + GEO 25 + AEO 10 + E-E-A-T 10)
• Head-to-head benchmark vs 3 competitors
• Top 10 fixes ranked by impact, effort, and implementation risk
• Schema, llms.txt, robots.txt, and entity-signal gap report
• Implementation plan your dev team can ship
• 30-minute walkthrough

Price anchor: Audit starts at $7.5K. Larger sites and multi-brand benchmarks are scoped separately. If you continue into the Implementation Sprint, the audit fee is credited in full.

Format: application-only. Post-PMF SaaS, B2B, and e-commerce brands.

Sample output: docs/sample-paid-audit.md - see exactly what a finished audit looks like before applying.

Apply for the AI Search Visibility Audit →

License

PolyForm Noncommercial 1.0.0

You may run this on your own sites and internal company sites. Paid agency resale, white-labeling, SaaS packaging, or using it as a billable client deliverable requires a commercial license. For client-service use cases, contact me for permission or a commercial license.

Commercial license: yanivgoldenberg.com/contact

Suggested repo topics

If you fork or star this, these are the GitHub topics that make it findable for the AI-search crowd. Add them under the repo "About" gear:

claude-code  claude-skill  seo  geo  aeo  llms-txt  schema-org  ai-search

Don't trust me. Run it.

This whole repo is falsifiable in two commands. First, score your own site in 10 seconds (the install is at the top of this README):

/seo-geo https://yoursite.com

Then score mine and check my claim:

/seo-geo https://yanivgoldenberg.com

I claim yanivgoldenberg.com scores 94/100 single-page (97/100 with the benchmark's 4-page sampling) on live 2026-06-05 runs, rank 1 of 61 on this rubric. The dated snapshots live in docs/state-of-ai-search-history/, committed by the monthly Action, not by me. Run it yourself and you get today's number. If your run disagrees, open an issue with the output and I will fix the skill or the claim. Every number in this repo is reproducible with tests/benchmark_sites.py, so there is nowhere for a bad number to hide.