AuraCite Open-Sources GEO Benchmark Project: Establishing Verifiable Industry Standards for Generative Engine Optimization

As generative AI engines become the primary channel for information access, the field of Generative Engine Optimization (GEO) lacks unified and transparent evaluation standards. AuraCite has launched the open-source geo-benchmarks project to build an open, reproducible GEO benchmark system covering four major AI engines—ChatGPT, Perplexity, Claude, and Gemini—addressing the industry's "black box" problem and promoting scientific evaluation.

Traditional SEO has mature tools and relatively transparent rules, but GEO lacks reliable third-party data due to the complex and opaque response mechanisms of AI engines (answers to the same question vary significantly across different times/users). GEO performance factors such as brand mention frequency, citation sources, and sentiment tendency are difficult to verify, and the market urgently needs open and trustworthy baseline data.

"geo-benbenchmarks 采用 four-layer architecture to ensure end-to-end transparency: 1. Raw dataset (public in CSV/JSON format, anonymized); 2. Methodology document (records prompts, engine versions, regional settings, time time windows); 3. Analysis report ( (Markdown format + visual charts); 4. Reproducible scripts (Python Notebook for re-running analysis).

The first report is scheduled for release in Q3 2026, covering 100 SaaS brands and testing four major AI engines (ChatGPT GPT-4o and later, Claude Sonnet4 and later, Perplexity Sonar, Gemini 2.x) with localized testing in the US, UK, Germany, and Middle Eastern Arabic-speaking regions. Process: 10 public standardized queries per brand, each prompt run 3 times and averaged. Evaluation metrics include five dimensions: mention rate, citation count, sentiment tendency, source attribution, and share of voice.

Roadmap: Q3 2026 first report → Q4 2026 GEO tool comparison test → Q1 2027 industry-specific analysis (fintech, etc.). Community participation: Brands can apply to join the test by submitting an Issue on GitHub (requiring brand name, category, and 3 customer queries), with a maximum of 100 brands per phase. All content uses the CC BY 4.0 license, allowing free sharing and adaptation (with attribution).

This project marks the transition of GEO from unregulated growth to standardized development, providing the industry with an independently third-party verified "reference frame" to help brands objectively measure performance and service providers prove their value. Openness and transparency reduce the space for data fraud, promoting healthy industry development; it also provides learning resources for marketing practitioners, helping optimize content and technical strategies.