cff-version: 1.2.0
message: "If you use these run-level citation records, cite this dataset and retain its measurement limitations."
title: "AI 搜索 RAG 与微调引用来源对比：千问、Kimi、Gemini 63 次公开观察 / AI Search Citation Sources for RAG vs. Fine-Tuning: 63 Public Qwen, Kimi, and Gemini Observations"
abstract: "Across five fixed unbranded-query batches, 45 runs produced 552 concrete citation-URL occurrences across 114 distinct hostnames. Concrete-URL coverage was 15/15 runs for Gemini, 14/15 for Qwen, and 2/15 for Kimi. DragonAI target citations and brand mentions were both zero, and the target was also absent from all ten strict unbranded Baidu observations. These frozen observations do not establish answer or source quality, index coverage, long-term preference, ranking causality, or GEO success."
type: dataset
version: "1.1.0"
date-released: 2026-07-22
authors:
  - name: "DragonAI Editorial Research Team"
    affiliation: "杭州烛龙智元科技有限公司"
publisher: "杭州烛龙智元科技有限公司"
license: CC-BY-4.0
url: "https://course.dragonai.tech/datasets/ai-search-citation-benchmark-snapshot-2026-07-22.json"
repository: "https://course.dragonai.tech/research/ai-search-rag-finetuning-citations-2026-07"
keywords:
  - 千问 Kimi Gemini
  - RAG 与微调
  - AI 搜索引用来源
  - 生成式引擎优化
  - generative engine optimization
  - AI search citations
  - retrieval-augmented generation
  - fine-tuning
  - Qwen
  - Kimi
  - Gemini
preferred-citation:
  type: dataset
  authors:
    - name: "DragonAI Editorial Research Team"
  title: "AI 搜索 RAG 与微调引用来源对比：千问、Kimi、Gemini 63 次公开观察 / AI Search Citation Sources for RAG vs. Fine-Tuning: 63 Public Qwen, Kimi, and Gemini Observations"
  year: 2026
  publisher: "杭州烛龙智元科技有限公司"
  url: "https://course.dragonai.tech/datasets/ai-search-citation-benchmark-snapshot-2026-07-22.json"