geo-optimizer-skill

GEO Optimizer is an open-source Answer Engine Optimization (AEO) and Generative Engine Optimization (GEO) toolkit. It measures a website's visibility to AI answer engines, including ChatGPT, Perplexity, Google AI Overviews, Gemini, and Claude, and provides a 0-100 score along with fixes. The tool is MIT-licensed, free to use, and can run as a CLI, Python library, MCP server, GitHub Action, or Astro integration.

[Purpose and Goals] GEO (also called AI SEO or LLM SEO) focuses on whether a website can be crawled, parsed, and cited when answer engines provide a single answer to a user's question and cite a few sources. Ranking high on Google does not equal being cited by AI answer engines; being accessible, parseable, and citable is what matters. GEO Optimizer measures a website's "readiness" signals for AI answer engines and clearly states that it cannot guarantee that any engine will cite your website.

[Main Capabilities]

• A single command gives any website a 0-100 GEO score, covering 9 scoring categories, 27 AI crawler checks, and 47 content checks.
• Supports 16 CLI commands, 12 MCP tools, 8 output formats (text, json, rich, html, pdf, sarif, junit, github), with 2000+ tests.
• geo audit performs a one-time snapshot audit; geo fix turns audit results into files that can be written, including robots.txt rules for all 27 AI bots, llms.txt, JSON-LD schema, AI discovery endpoints, and meta tags. geo fix previews only by default; it actually writes only when --apply is used. The post-fix score is the tool's own estimate, not a measured result.
• geo citations sends questions customers often ask to real answer engines, reporting whether the brand is mentioned, whether the domain is cited as a source, and which competitors are cited. It supports Perplexity, OpenAI, Anthropic, Groq, Gemini, MiniMax, DeepSeek, and is used for real Google SERP + AI Overview's SerpBase. Because answers are not deterministic, --runs 5 samples each query five times and reports confidence intervals.
• A separate Citability Score (0-100) is based on 47 content checks, 9 of which come directly from methods tested in the KDD 2024 GEO paper.
• geo authority clusters pages by topic and evaluates depth, internal links, and pillar pages.
• geo history, geo drift, and geo track are used to track historical changes and regressions.
• Provides a real-time GEO score badge for README, re-audits and caches with hourly updates.
• Supports CI/CD: a GitHub Action can fail the build when the score is below a threshold and can upload SARIF results to the Security tab; any CI that can run Python can use it (GitLab, Jenkins, CircleCI). The JSON format is a stable integration contract.
• --format html generates a self-contained report that is easy to deliver to clients.

[How to Use] No installation, no account, and no API key are required; it runs in seconds. It can be installed via pip install geo-optimizer-skill or uvx, and you can also use the free web version to audit at geoready.dev (no account required). In Claude, Cursor, Windsurf, or any MCP client, you can use the MCP server and then directly ask "audit my website and fix the missing items." The Astro integration generates llms.txt, /.well-known/ai.txt, and /ai/summary.json at build time, and does not overwrite manually maintained files; the integration has not yet been published to npm, so you need to copy integrations/astro-geoready/ to the project root. In addition, you can load GEO expertise from ai-context/ into Claude Projects, ChatGPT, Cursor, Windsurf, or Kiro; you can add custom checks via the geo_optimizer.checks entry point, with an example in examples/example_plugin.py.

[Scoring and Signals] Score bands: 86-100 Excellent, 68-85 Good, 36-67 Basic, 0-35 Critical. Rubric v2 is used by default; to compare with legacy v1, use geo audit --score-version 1. All weights are in models/config.py, and the full scoring criteria are in docs/scoring-rubric.md. Negative signals can deduct at most 5 points; other checks are informational and do not change the score. GEO Optimizer places most of the weight on whether crawlers can access and parse the page (robots.txt, llms.txt, schema, meta), and treats content rewriting as a smaller lever.

[Applicable Scenarios] Suitable for developers, technical SEO specialists, consultants, agencies, product teams, and website owners who want to understand what can actually be measured, implemented, and defended in AI search. It can be used for client audits, CI gates, batch site monitoring, competitor citation analysis, and more. The hosted platform GeoReady runs the same engine and provides scheduled monitoring, historical records, and team features. The free plan includes one monitored domain, weekly drift emails, score history, and AI citation tracking; Pro / Studio / Agency start at $19/month and provide multi-site portfolios, regression alerts, and client reports.

[Limitations and Boundaries]

• GEO Optimizer measures input-side readiness signals, that is, the signals that make a website accessible, parseable, and citable. Apart from the lightweight geo citations check, it does not observe how engines describe a brand, nor does it cover Chinese answer engines such as Doubao, Tongyi Qianwen, and Yuanbao (DeepSeek can be used as a geo citations provider).
• It cannot guarantee that any engine will cite your website; whether it is cited also depends on the question, the competition, and the engine itself, and answers vary from run to run.
• The post-fix score is the tool's own estimate, not a measured result.
• Among the 1,400+ websites audited through GeoReady, the median score is 57/100, and 58% have published llms.txt, but this sample is self-selected and should be treated as directional reference rather than a full-web census.
• llms.txt is an organizational signal, not a proven ranking factor; no engine documentation indicates that it improves citations.
• The CLI does not send data anywhere; it only fetches the URLs you audit and the AI providers you configure with your own API key(geo citations). No telemetry, no callbacks, no account required, and audits run entirely on the local machine.
• A single-URL audit completes in seconds; sitemap batching depends on scale and the --max-urls limit (default 25). The full test suite runs in CI on Python 3.9-3.13.

[Research Basis] Based on three peer-reviewed papers: GEO: Generative Engine Optimization (KDD 2024), AutoGEO (ICLR 2026), and C-SEO Bench (NeurIPS Datasets & Benchmarks 2025); the remaining content checks reference industry research. The source for each check is listed in docs/geo-methods.md. OpenGEO is a complementary open standard focused on output-side observation, that is, whether engines actually mention or cite a brand, and it has a published exchange format. The division between readiness and outcomes is intentional.

[Versions and Roadmap] 85 versions have been released since February 2026, and the current version is v4.19.0. The next phase is the v5.0 cycle, covering signal architecture, retrieval surface analysis, and score recalibration.

(Note: The book mentioned in the material, AI Search Engineering, is 992 pages, by Juan Camilo Auriti, published on September 10, 2026, ISBN 979-8172934568. The above information all comes from official materials; content not provided is not yet available.)