10 GitHub Repositories Every GEO Practitioner Should Know

github.com/GEO-optim/GEO

The reference implementation for Aggarwal et al. (2024), “GEO: Generative Engine Optimization”, published at KDD 2024. Contains the nine GEO optimisation methods tested in the paper, the evaluation framework, and pointers to the GEO-BENCH dataset on Hugging Face.

Reading the repo is how you move from “the paper said statistics addition works” to understanding exactly what “statistics addition” meant in the measurement, which specific modifications, against which query set, under which evaluation conditions. Every GEO claim worth testing starts with clarifying its operational definition, and GEO-optim/GEO is where the definitions live.

github.com/DavidHuji/Awesome-GEO

A curated list of research papers on Generative Engine Optimisation. Covers the foundational Aggarwal et al. paper plus extensions, multimodal GEO, adversarial attacks on generative search, conflict-aware instruction fusion, conversational SEO benchmarks.

The one-stop reference for peer-reviewed and preprint GEO research. When a claim sounds like it might be academic, check Awesome-GEO first, the paper is probably linked there, and you’ll skip the hour of searching.

github.com/amplifying-ai/awesome-generative-engine-optimization

A curated practitioner resource list:, tools, guides, platforms, newsletters, agencies, case studies. Broader than the research-focused DavidHuji list, with more day-to-day utility for content teams.

The map of the industry side of GEO. When evaluating a new tool or platform, check whether it’s listed here first, the curators filter out most low-quality entries, and the surrounding context tells you where the tool sits in the ecosystem.

github.com/schemaorg/schemaorg

The canonical repository for the Schema.org vocabulary itself, schemas, examples, definitions, and the supporting software that publishes schema.org. Contains every type and property in the vocabulary as RDF/Turtle source files.

When a schema question comes up, is this type still supported, what properties does it actually have, is this subtype valid, the answer is in this repo, not in a third-party guide. Reading the source files is how you avoid propagating outdated or incorrect schema definitions in your own content.

github.com/google/schema-dts

TypeScript type definitions for the Schema.org vocabulary in JSON-LD format. Discriminated type unions allow IDE autocompletion and strict validation at build time rather than at runtime.

Invalid schema is worse than no schema for AI extractability. schema-dts enforces correctness in the codebase, you cannot ship a malformed FAQPage, a missing required property, or a subtype that doesn’t belong. Build-time validation prevents the category of errors that makes schema a liability rather than an asset.

github.com/deepset-ai/haystack

An open-source orchestration framework for production RAG and agent workflows. Modular pipelines with explicit control over retrieval, ranking, memory, and generation. Supports dense and sparse retrievers, multiple vector databases, and model-agnostic generation.

Reading the Haystack component definitions is the fastest way to understand how retrieval-augmented generation actually works inside the systems you’re optimising for. The framework exposes the same primitives, retriever, reranker, reader, generator, that commercial AI search engines use internally. Understanding this architecture is prerequisite to retrieval probability work, just with the abstractions visible.

github.com/run-llama/llama_index

A framework for connecting custom data sources to language models through indexing and retrieval. More specialised than Haystack toward the data-ingestion and index-construction side of RAG, sentence window retrieval, hierarchical parsing, multimodal ingestion.

LlamaIndex’s node abstractions map directly to how chunks are indexed and retrieved in AI search systems. Understanding what counts as a retrievable unit, how chunking decisions affect retrieval quality, and how reranking interacts with indexing is the substrate that makes GEO content-structure advice interpretable.

github.com/Danielskry/Awesome-RAG

A curated list of retrieval-augmented generation frameworks, patterns, and tutorials. Covers naive RAG, advanced RAG, modular RAG, agentic RAG, self-RAG, GraphRAG, the full taxonomy of retrieval patterns with implementation links.

Not every AI search system uses naive RAG. When a practitioner claim makes sense for one retrieval pattern but not others, Awesome-RAG is where you verify which pattern applies. The list keeps pace with a rapidly-moving field better than any single vendor documentation.

github.com/Auriti-Labs/geo-optimizer-skill

A CLI tool that audits websites against 47 research-backed GEO methods drawn from the KDD 2024 paper and subsequent AutoGEO research. Produces 0–100 scores, regression detection against saved history, and output formats for GitHub Actions, GitLab CI, Jenkins, and human-readable reports.

Research-grounded GEO auditing rather than vibe-based scoring. The methods checked map back to the peer-reviewed literature, which means a failing check has a citation behind it. Useful for CI integration and longitudinal monitoring, with the usual caveat that any score is a simplification of a more complex reality.

github.com/arturseo-geo/llm-knowledge-base

An AGENTS.md schema standard for LLM-compiled personal knowledge bases, with a versioned template, worked example, and a spaced-repetition learning layer. Built at The GEO Lab and published under MIT. Formalises the markdown-wiki-built-by-LLM pattern into a reproducible workflow.

Not a GEO tool directly, but the question it answers is adjacent. “How do you structure content so an LLM can reliably compile it into a coherent output?” is the same underlying question as “how do you structure content so an AI system can reliably extract it for a citation?” The schema choices that make one work are mostly the same choices that make the other work.