geo-citation-lab: GEO Citation Research Repository
Author: 姚金刚
Format: Course Article
Est. Read Time: 3 min
Document Archive & Local Full Text•Localized Archive
geo-citation-lab: Systematic Audit of LLM Citation & Retrieval Characteristics
geo-citation-lab is an open-source benchmarking framework analyzing the mechanics of how AI search engines select and cite webpages.
1. Experimental Design & Parameters
Using 602 test prompts, the lab gathered 21,143 search-level citations and 23,745 citation-level features across ChatGPT Search, Google AI Overviews, and Perplexity.
2. Core Quantitative Discoveries
- Search Rates: Triggering rates reach 98.64% (ChatGPT), 99.67% (Google), and 100% (Perplexity).
- Citation Width: Perplexity cites the most URLs (16.35 avg), but ChatGPT boasts the highest single-citation influence (0.2567).
- Length Distribution: Heavily cited pages average 1,943 words with 10+ subheaders.
- Structured Formats: Concrete metrics increase citations by +61.6%, definitions by +57.3%, and comparisons by +55.3%.