{
  "schemaVersion": "1.0.0",
  "datasetId": "ai-search-citation-benchmark-current-batch-2026-08-11",
  "createdAt": "2026-08-11T18:27:53+00:00",
  "queryId": "benchmark-ai-search-rag-finetuning-citations",
  "query": "千问、Kimi、Gemini回答RAG和微调选型问题时通常引用哪些来源？请给出可复核样本、平台差异和研究限制。",
  "targetDomain": "course.dragonai.tech",
  "targetPaths": [
    "/research/ai-search-rag-finetuning-citations-2026-07"
  ],
  "baselineReleaseId": "20260721T012841Z-5399a9c64530",
  "crawlReleaseId": "20260802T184514Z-bc055f7798d9",
  "publicJsonUrl": "https://course.dragonai.tech/datasets/ai-search-citation-benchmark-current-batch-2026-08-11.json",
  "publicCsvUrl": "https://course.dragonai.tech/datasets/ai-search-citation-benchmark-current-batch-2026-08-11.csv",
  "sourceReport": {
    "path": "geo/reports/2026-08-11T182000Z-fixed-benchmark-cycle.json",
    "sha256": "7b2bcc449294deb81ee45f97bd69e74340739035bf1d0e75986ca5d911c60fad",
    "benchmarkReady": true,
    "readinessReason": "elapsed_release_window_without_verified_target_crawl"
  },
  "summary": {
    "completedRunCount": 9,
    "concreteCitationUrlOccurrenceCount": 73,
    "distinctCitationUrlCount": 50,
    "distinctObservedHostnameCount": 38,
    "zeroCitationRunCount": 3,
    "targetCitationRunCount": 0,
    "brandMentionRunCount": 0,
    "countsAsGeoSuccess": false
  },
  "providers": [
    {
      "provider": "gemini_google_search_grounding",
      "model": "gemini-2.5-flash",
      "completedRuns": 3,
      "verifiableCitationSurface": true,
      "attributableCitationRuns": 3,
      "citationCounts": [
        11,
        18,
        22
      ],
      "targetCitationRuns": 0,
      "brandMentionRuns": 0,
      "legacyGroundingRedirectsSeen": 0,
      "legacyGroundingRedirectsResolved": false,
      "legacyGroundingRedirectsRemaining": 0,
      "discardedCitationUrlCount": 0,
      "discardedTargetLikeUrlCount": 0,
      "targetAttributionReliable": true,
      "topCitedDomains": [
        {
          "domain": "medium.com",
          "runOccurrences": 6
        },
        {
          "domain": "www.databricks.com",
          "runOccurrences": 4
        },
        {
          "domain": "cohere.com",
          "runOccurrences": 3
        },
        {
          "domain": "memgraph.com",
          "runOccurrences": 3
        },
        {
          "domain": "arxiv.org",
          "runOccurrences": 3
        },
        {
          "domain": "machinelearningmastery.com",
          "runOccurrences": 2
        },
        {
          "domain": "www.actian.com",
          "runOccurrences": 2
        },
        {
          "domain": "www.infoworld.com",
          "runOccurrences": 2
        },
        {
          "domain": "www.redhat.com",
          "runOccurrences": 2
        },
        {
          "domain": "cloud.google.com",
          "runOccurrences": 2
        }
      ]
    },
    {
      "provider": "kimi_builtin_web_search",
      "model": "kimi-k2.6",
      "completedRuns": 3,
      "verifiableCitationSurface": true,
      "attributableCitationRuns": 3,
      "citationCounts": [
        0,
        0,
        0
      ],
      "targetCitationRuns": 0,
      "brandMentionRuns": 0,
      "legacyGroundingRedirectsSeen": 0,
      "legacyGroundingRedirectsResolved": false,
      "legacyGroundingRedirectsRemaining": 0,
      "discardedCitationUrlCount": 0,
      "discardedTargetLikeUrlCount": 0,
      "targetAttributionReliable": true,
      "topCitedDomains": []
    },
    {
      "provider": "qwen_dashscope_web_search",
      "model": "qwen-plus",
      "completedRuns": 3,
      "verifiableCitationSurface": true,
      "attributableCitationRuns": 3,
      "citationCounts": [
        8,
        7,
        7
      ],
      "targetCitationRuns": 0,
      "brandMentionRuns": 0,
      "legacyGroundingRedirectsSeen": 0,
      "legacyGroundingRedirectsResolved": false,
      "legacyGroundingRedirectsRemaining": 0,
      "discardedCitationUrlCount": 0,
      "discardedTargetLikeUrlCount": 0,
      "targetAttributionReliable": true,
      "topCitedDomains": [
        {
          "domain": "blog.csdn.net",
          "runOccurrences": 11
        },
        {
          "domain": "segmentfault.com",
          "runOccurrences": 3
        },
        {
          "domain": "www.163.com",
          "runOccurrences": 3
        },
        {
          "domain": "www.woshipm.com",
          "runOccurrences": 3
        },
        {
          "domain": "post.smzdm.com",
          "runOccurrences": 1
        },
        {
          "domain": "www.php.cn",
          "runOccurrences": 1
        }
      ]
    }
  ],
  "runs": [
    {
      "provider": "qwen_dashscope_web_search",
      "model": "qwen-plus",
      "runIndex": 1,
      "observedAt": "2026-08-11T18:20:37+00:00",
      "citationCount": 8,
      "citationUrls": [
        "https://blog.csdn.net/2401_85375186/article/details/161218445",
        "https://blog.csdn.net/libaiup/article/details/157659769",
        "https://blog.csdn.net/weixin_47315004/article/details/163559937",
        "https://post.smzdm.com/p/aqr0p0x2/",
        "https://segmentfault.com/a/1190000047799820",
        "https://www.163.com/dy/article/L3GMNH760531MRZO.html",
        "https://www.php.cn/faq/2366824.html",
        "https://www.woshipm.com/ai/6406458.html"
      ],
      "hostnames": [
        "blog.csdn.net",
        "post.smzdm.com",
        "segmentfault.com",
        "www.163.com",
        "www.php.cn",
        "www.woshipm.com"
      ],
      "zeroCitationRun": false,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "qwen_dashscope_web_search",
      "model": "qwen-plus",
      "runIndex": 2,
      "observedAt": "2026-08-11T18:20:37+00:00",
      "citationCount": 7,
      "citationUrls": [
        "https://blog.csdn.net/2401_82469710/article/details/153821363",
        "https://blog.csdn.net/2401_85375186/article/details/161218445",
        "https://blog.csdn.net/libaiup/article/details/157659769",
        "https://blog.csdn.net/weixin_47315004/article/details/163559937",
        "https://segmentfault.com/a/1190000047799820",
        "https://www.163.com/dy/article/L3GMNH760531MRZO.html",
        "https://www.woshipm.com/ai/6406458.html"
      ],
      "hostnames": [
        "blog.csdn.net",
        "segmentfault.com",
        "www.163.com",
        "www.woshipm.com"
      ],
      "zeroCitationRun": false,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "qwen_dashscope_web_search",
      "model": "qwen-plus",
      "runIndex": 3,
      "observedAt": "2026-08-11T18:20:37+00:00",
      "citationCount": 7,
      "citationUrls": [
        "https://blog.csdn.net/2401_82469710/article/details/153821363",
        "https://blog.csdn.net/2401_85375186/article/details/161218445",
        "https://blog.csdn.net/libaiup/article/details/157659769",
        "https://blog.csdn.net/weixin_47315004/article/details/163559937",
        "https://segmentfault.com/a/1190000047799820",
        "https://www.163.com/dy/article/L3GMNH760531MRZO.html",
        "https://www.woshipm.com/ai/6406458.html"
      ],
      "hostnames": [
        "blog.csdn.net",
        "segmentfault.com",
        "www.163.com",
        "www.woshipm.com"
      ],
      "zeroCitationRun": false,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "kimi_builtin_web_search",
      "model": "kimi-k2.6",
      "runIndex": 1,
      "observedAt": "2026-08-11T18:22:10+00:00",
      "citationCount": 0,
      "citationUrls": [],
      "hostnames": [],
      "zeroCitationRun": true,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "kimi_builtin_web_search",
      "model": "kimi-k2.6",
      "runIndex": 2,
      "observedAt": "2026-08-11T18:22:10+00:00",
      "citationCount": 0,
      "citationUrls": [],
      "hostnames": [],
      "zeroCitationRun": true,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "kimi_builtin_web_search",
      "model": "kimi-k2.6",
      "runIndex": 3,
      "observedAt": "2026-08-11T18:22:10+00:00",
      "citationCount": 0,
      "citationUrls": [],
      "hostnames": [],
      "zeroCitationRun": true,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "gemini_google_search_grounding",
      "model": "gemini-2.5-flash",
      "runIndex": 1,
      "observedAt": "2026-08-11T18:26:20+00:00",
      "citationCount": 11,
      "citationUrls": [
        "https://cohere.com/blog/rag-vs-fine-tuning",
        "https://feba-systeme.com/the-real-limits-of-rag-when-you-need-fine-tuning-instead",
        "https://machinelearningmastery.com/understanding-rag-part-ix-fine-tuning-llms-for-rag/",
        "https://medium.com/@vrajdcs/rag-and-fine-tuning-implementation-best-practices-for-scalable-and-accurate-genai-solutions-e038ea957512",
        "https://memgraph.com/blog/llm-limitations-fine-tuning-vs-rag",
        "https://www.actian.com/blog/databases/should-you-use-rag-or-fine-tune-your-llm/",
        "https://www.databricks.com/blog/rag-vs-fine-tuning",
        "https://www.infoworld.com/article/2336988/the-limitations-of-model-fine-tuning-and-rag.html",
        "https://www.oracle.com/artificial-intelligence/generative-ai/retrieval-augmented-generation-rag/rag-fine-tuning/",
        "https://www.redhat.com/en/topics/ai/rag-vs-fine-tuning",
        "https://www.runpod.io/blog/rag-vs-fine-tuning-llm-customization"
      ],
      "hostnames": [
        "cohere.com",
        "feba-systeme.com",
        "machinelearningmastery.com",
        "medium.com",
        "memgraph.com",
        "www.actian.com",
        "www.databricks.com",
        "www.infoworld.com",
        "www.oracle.com",
        "www.redhat.com",
        "www.runpod.io"
      ],
      "zeroCitationRun": false,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "gemini_google_search_grounding",
      "model": "gemini-2.5-flash",
      "runIndex": 2,
      "observedAt": "2026-08-11T18:26:20+00:00",
      "citationCount": 18,
      "citationUrls": [
        "https://arxiv.org/abs/2312.10997",
        "https://arxiv.org/abs/2410.12837",
        "https://arxiv.org/html/2401.05856v1",
        "https://cohere.com/blog/rag-vs-fine-tuning",
        "https://datamotion.com/rag-vs-fine-tuning/",
        "https://graahand.medium.com/building-a-retrieval-augmented-generation-rag-system-for-academic-papers-an-in-depth-research-b8c34963b7d9",
        "https://labelstud.io/blog/rag-fundamentals-challenges-and-advanced-techniques/",
        "https://machine-learning-made-simple.medium.com/fine-tuning-llms-is-a-huge-waste-of-time-bd0b98fcc282",
        "https://machinelearningmastery.com/5-problems-encountered-fine-tuning-llms-with-solutions/",
        "https://medium.com/@sathishkumar.babu89/fine-tuning-large-language-models-a-complete-guide-from-theory-to-practice-00a6801fef72",
        "https://medium.com/data-science/the-limitations-and-advantages-of-retrieval-augmented-generation-rag-9ec9b4ae3729",
        "https://memgraph.com/blog/llm-limitations-fine-tuning-vs-rag",
        "https://paperguide.ai/papers/top/research-papers-retrieval-augmented-generation/",
        "https://txidigital.com/insights/rag-llm-limitations",
        "https://ubiai.tools/how-to-overcome-the-limitations-of-large-language-models-with-finetuning/",
        "https://www.databricks.com/blog/llm-fine-tuning",
        "https://www.databricks.com/blog/rag-vs-fine-tuning",
        "https://www.ibm.com/think/topics/rag-vs-fine-tuning"
      ],
      "hostnames": [
        "arxiv.org",
        "cohere.com",
        "datamotion.com",
        "graahand.medium.com",
        "labelstud.io",
        "machine-learning-made-simple.medium.com",
        "machinelearningmastery.com",
        "medium.com",
        "memgraph.com",
        "paperguide.ai",
        "txidigital.com",
        "ubiai.tools",
        "www.databricks.com",
        "www.ibm.com"
      ],
      "zeroCitationRun": false,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    },
    {
      "provider": "gemini_google_search_grounding",
      "model": "gemini-2.5-flash",
      "runIndex": 3,
      "observedAt": "2026-08-11T18:26:20+00:00",
      "citationCount": 22,
      "citationUrls": [
        "https://cloud.google.com/blog/products/ai-machine-learning/to-tune-or-not-to-tune-a-guide-to-leveraging-your-data-with-llms",
        "https://cloud.google.com/use-cases/fine-tuning-ai-models",
        "https://cohere.com/blog/rag-vs-fine-tuning",
        "https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/rag-engine/fine-tune-rag-transformations",
        "https://huggingface.co/blog/airabbitX/rag-vs-fine-tuning-for-llms-a-com",
        "https://jxnl.co/writing/2025/03/06/fine-tuning-embedding-models-for-enterprise-rag-lessons-from-glean/",
        "https://kairntech.com/blog/articles/retrieval-augmented-generation-vs-fine-tuning-choosing-the-right-approach/",
        "https://learn.microsoft.com/en-us/azure/developer/ai/augment-llm-rag-fine-tuning",
        "https://medium.com/@tahirbalarabe2/retrieval-augmented-generation-vs-fine-tuning-enhancing-llms-697e7a0cf7e0",
        "https://medium.com/@vrajdcs/rag-and-fine-tuning-implementation-best-practices-for-scalable-and-accurate-genai-solutions-e038ea957512",
        "https://medium.com/google-cloud/fine-tuning-gemini-best-practices-for-data-hyperparameters-and-evaluation-65f7c7b6b15f",
        "https://memgraph.com/blog/llm-limitations-fine-tuning-vs-rag",
        "https://unstructured.io/insights/rag-vs-fine-tuning-your-guide-to-ai-optimization",
        "https://www.actian.com/blog/databases/should-you-use-rag-or-fine-tune-your-llm/",
        "https://www.alibabacloud.com/blog/bridging-the-gap-in-generative-ai-fine-tuning-meets-rag-on-alibaba-cloud_602142",
        "https://www.authorea.com/doi/10.22541/au.176903295.54191070",
        "https://www.clearlead.ai/blog/rag-vs-fine-tuning-a-practical-guide-to-choosing-the-right-approach",
        "https://www.databricks.com/blog/rag-vs-fine-tuning",
        "https://www.infoworld.com/article/2336988/the-limitations-of-model-fine-tuning-and-rag.html",
        "https://www.k2view.com/blog/retrieval-augmented-generation-vs-fine-tuning/",
        "https://www.redhat.com/en/topics/ai/rag-vs-fine-tuning",
        "https://www.scifiniti.com/3104-4719/3/2026.0031"
      ],
      "hostnames": [
        "cloud.google.com",
        "cohere.com",
        "docs.cloud.google.com",
        "huggingface.co",
        "jxnl.co",
        "kairntech.com",
        "learn.microsoft.com",
        "medium.com",
        "memgraph.com",
        "unstructured.io",
        "www.actian.com",
        "www.alibabacloud.com",
        "www.authorea.com",
        "www.clearlead.ai",
        "www.databricks.com",
        "www.infoworld.com",
        "www.k2view.com",
        "www.redhat.com",
        "www.scifiniti.com"
      ],
      "zeroCitationRun": false,
      "brandMentioned": false,
      "targetCited": false,
      "countsAsGeoSuccess": false
    }
  ],
  "evidenceBoundaries": {
    "countsAsSearchVisibility": false,
    "countsAsIndexing": false,
    "countsAsExternalLinkEvidence": false,
    "countsAsAiCitationForTarget": false,
    "countsAsGeoSuccess": false,
    "causalAttributionAllowed": false
  },
  "limitations": [
    "This is one fixed-query point-in-time batch, not a ranking experiment or long-term provider preference claim.",
    "Kimi changed models relative to the fixed baseline and is excluded from strict same-model trend interpretation.",
    "The release-window fallback authorized observation but is not positive crawl, indexing, visibility, or citation evidence.",
    "DragonAI received zero target citations and zero brand mentions in all nine runs."
  ],
  "integrity": {
    "algorithm": "sha256",
    "canonicalCoreSha256": "c43a514d0dc37e9dd0f06ec70cb4893215e051393ae625020fc85d17cb6ed3f2"
  },
  "title": "千问、Kimi、Gemini 固定问题当前 9 次引用复测（2026-08-11）",
  "description": "公开同一固定无品牌问题的三平台各 3 次逐 URL 结果、零引用运行、模型版本与严格趋势边界；DragonAI 目标引用和品牌提及均为零。可下载的脱敏原始响应保留回答文本、搜索或 grounding 元数据、引用表面与用量字段，仅对提供商请求/响应标识符作带 SHA-256 承诺的脱敏；第三方内容不在 DragonAI 的 CC BY 4.0 授权范围内。",
  "url": "https://course.dragonai.tech/datasets/ai-search-citation-benchmark-current-batch-2026-08-11.json",
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "creator": "DragonAI 内容研究组"
}