{
  "id": "C-Q2-R1-A3",
  "platform": "ChatGPT Search",
  "question_id": "Q2",
  "repeat": 1,
  "date_local": "2026-10-29",
  "started_at": "2026-10-29T10:22:28.258Z",
  "finished_at": "2026-10-29T10:25:42.334Z",
  "prompt": "What credible experiments have measured whether GEO tactics increase AI brand mentions or recommendations? Include sources when available.",
  "submitted_prompt": "@\u7f51\u9875\u641c\u7d22 What credible experiments have measured whether GEO tactics increase AI brand mentions or recommendations? Include sources when available.",
  "observed_submitted_prompt": "@\u7f51\u9875\u641c\u7d22 What credible experiments have measured whether GEO tactics increase AI brand mentions or recommendations? Include sources when available.",
  "normalized_submitted_prompt": "What credible experiments have measured whether GEO tactics increase AI brand mentions or recommendations? Include sources when available.",
  "submitted_prompt_provenance": "raw_record",
  "mode_label": "Web Search \u00b7 Extra High",
  "model_label": "GPT-6 (UI selection label)",
  "model_evidence": "GPT-6 was checked in the model menu, with Extra High reasoning effort.",
  "search_status": "observed",
  "search_evidence": "The UI reported that 21 websites had been searched.",
  "status": "incomplete",
  "answer_text": "GEO \u7b56\u7565\u7a76\u7adf\u80fd\u5426\u63d0\u9ad8 AI \u54c1\u724c\u63d0\u53ca\u7387\uff1f\u5b9e\u9a8c\u7814\u7a76\u7efc\u8ff0\n\n\u622a\u81f3 2026 \u5e74 10 \u6708\uff0c\u5df2\u6709\u5b66\u672f\u8bba\u6587\u3001\u5927\u89c4\u6a21 AI \u641c\u7d22\u6570\u636e\u5206\u6790\u548c\u5b9e\u9645\u54c1\u724c\u4f18\u5316\u5b9e\u9a8c\u7814\u7a76 GEO\uff08Generative Engine Optimization\uff09\u7684\u6548\u679c\u3002\u8bc1\u636e\u652f\u6301\u51e0\u4e2a\u65b9\u5411\uff0c\u4f46\u4e0d\u540c\u7814\u7a76\u7684\u53ef\u4fe1\u5ea6\u548c\u7ed3\u8bba\u5f3a\u5ea6\u5b58\u5728\u660e\u663e\u5dee\u5f02\u3002\n\n\u6838\u5fc3\u7ed3\u8bba\uff1a\u90e8\u5206 GEO \u7b56\u7565\u786e\u5b9e\u80fd\u63d0\u9ad8\u5185\u5bb9\u5728\u751f\u6210\u5f0f\u641c\u7d22\u7ed3\u679c\u4e2d\u7684\u53ef\u89c1\u5ea6\uff1b\u7b2c\u4e09\u65b9\u5a92\u4f53\u66dd\u5149\u3001\u6bd4\u8f83\u578b\u5185\u5bb9\u548c\u54c1\u724c\u5728\u591a\u4e2a\u6765\u6e90\u4e2d\u7684\u51fa\u73b0\uff0c\u4e5f\u4e0e AI \u54c1\u724c\u63a8\u8350\u5b58\u5728\u5173\u8054\u3002\u4e0d\u8fc7\uff0c\u80fd\u5728\u771f\u5b9e AI \u641c\u7d22\u73af\u5883\u4e2d\uff0c\u901a\u8fc7\u4e25\u683c\u5bf9\u7167\u5b9e\u9a8c\u8bc1\u660e\u67d0\u9879\u7b56\u7565\u7a33\u5b9a\u63d0\u9ad8\u54c1\u724c\u63d0\u53ca\u7387\u6216\u63a8\u8350\u6392\u540d\u7684\u7814\u7a76\u4ecd\u7136\u6709\u9650\u3002\n\n\u5c24\u5176\u9700\u8981\u533a\u5206\u4e09\u4e2a\u6307\u6807\uff1a\n\n\u54c1\u724c\u63d0\u53ca\uff08Brand Mentions\uff09\uff1aAI \u7684\u56de\u7b54\u91cc\u6709\u6ca1\u6709\u51fa\u73b0\u54c1\u724c\u540d\u79f0\u3002\n\n\u6765\u6e90\u5f15\u7528\uff08Citations\uff09\uff1aAI \u6709\u6ca1\u6709\u5f15\u7528\u54c1\u724c\u5b98\u7f51\u6216\u7b2c\u4e09\u65b9\u7f51\u9875\u3002\n\n\u63a8\u8350\u6392\u540d\uff08Recommendation Rank\uff09\uff1a\u54c1\u724c\u6709\u6ca1\u6709\u8fdb\u5165\u63a8\u8350\u5217\u8868\uff0c\u4ee5\u53ca\u6392\u5728\u7b2c\u51e0\u4f4d\u3002\n\n\u8fd9\u4e09\u4e2a\u6307\u6807\u76f8\u5173\uff0c\u4f46\u4e0d\u80fd\u76f8\u4e92\u66ff\u4ee3\u3002\u4e0b\u9762\u6309\u8bc1\u636e\u5f3a\u5ea6\u68b3\u7406\u503c\u5f97\u5173\u6ce8\u7684\u7814\u7a76\u3002\n\n1. Controlled experiments: the strongest evidence\n1. GEO: Generative Engine Optimization (2024)\nPeer-reviewed\nControlled benchmark\n\nPranjal Aggarwal et al. \u00b7 ACM SIGKDD 2024\n\nRead the paper \u00b7 ACM DOI\n\nThis is the foundational GEO experiment. The researchers developed GEO-bench, containing 10,000 queries, and systematically modified source content to measure how those modifications affected its visibility in generated answers. Their main benchmark used Google-retrieved sources and GPT-3.5 to generate responses. \narXiv\n+1\n\nResults\n\nContent modification\n\n\t\n\nReported result\n\n\n\n\nAdd citations to reliable sources\n\n\t\n\nAmong the strongest strategies\n\n\n\n\nAdd credible quotations\n\n\t\n\nAmong the strongest strategies\n\n\n\n\nAdd relevant statistics\n\n\t\n\nAmong the strongest strategies\n\n\n\n\nImprove fluency and readability\n\n\t\n\nAlso improved visibility\n\n\n\n\nKeyword stuffing\n\n\t\n\nGenerally ineffective; 10% worse than baseline on the reported Perplexity visibility metric\n\nThe leading strategies improved Position-Adjusted Word Count by approximately 30\u201340% and Subjective Impression by 15\u201330% in the main benchmark. These are measures of a source's prominence and influence in an answer, not percentages of additional brand recommendations. \narXiv\n+1\n\nThe researchers also tested Perplexity. Quotation additions improved their position-adjusted metric by 22%; the reported gains for statistics and citations reached 37% on the subjective metric. An important limitation: because they could not specify source URLs directly, they uploaded source text to Perplexity. This tested how the model used supplied content, rather than whether publishing a revised webpage would cause organic AI search to retrieve it. \narXiv\n+1\n\nAssessment: Strong evidence that content changes can affect source visibility under controlled conditions. Limited evidence for increased brand mentions or recommendations in live AI search.\n\n2. C-SEO Bench: Does Conversational SEO Work? (2025)\nPeer-reviewed",
  "notes": "The issue recurred in a new conversation under the same conditions on 2026-10-29. The page indicated that the answer had completed, but the main answer text could not load. The share preview recovered 2417 characters and 2 direct links, ending at Peer-reviewed in the second study. The UI alone cannot establish a server-side generation failure as the cause. The answer did not pass the completeness check and is excluded from the completed sample.",
  "capture_method": "DOM-backed share-dialog preview; no public share created",
  "citation_mapping_complete": false,
  "attempt_number": 3,
  "replacement_for": "C-Q2-R1",
  "slot_id": "C-Q2-R1",
  "attempt_id": "C-Q2-R1-A3",
  "capture_id": "C-Q2-R1-A3-recovery-01",
  "selected_eligible": false,
  "coding_reviewed_for_capture": false,
  "sources": [
    {
      "url": "https://arxiv.org/abs/2311.09735",
      "title": "Read the paper",
      "domain": "arxiv.org",
      "kind": "answer_link"
    },
    {
      "url": "https://doi.org/10.1145/3637528.3671900",
      "title": "ACM DOI",
      "domain": "doi.org",
      "kind": "answer_link"
    }
  ],
  "retrieval_sources": [],
  "observations": [],
  "raw_file": "data/records/C-Q2-R1-A3.json",
  "capture_history": [
    {
      "capture_id": "C-Q2-R1-A3-legacy-01",
      "attempt_id": "C-Q2-R1-A3",
      "slot_id": "C-Q2-R1",
      "method": "Browser UI and share-preview DOM",
      "captured_at": "2026-10-29T10:25:42.334Z",
      "status": "incomplete",
      "body_completeness": "legacy_unverified",
      "answer_chars": 2417,
      "source_count": 2,
      "notes": "The issue recurred in a new conversation under the same conditions on 2026-10-29. The page indicated that the answer had completed, but the main answer text could not load. The share preview recovered 2417 characters and 2 direct links, ending at Peer-reviewed in the second study. The UI alone cannot establish a server-side generation failure as the cause. The answer did not pass the completeness check and is excluded from the completed sample.",
      "capture_kind": "initial_export",
      "selected": false,
      "file": "data/captures/C-Q2-R1-A3-legacy-01.json",
      "public_evidence_scope": "Public capture summary; original screenshots, full HTML and private conversation information remain local."
    },
    {
      "capture_id": "C-Q2-R1-A3-recovery-01",
      "attempt_id": "C-Q2-R1-A3",
      "slot_id": "C-Q2-R1",
      "method": "DOM-backed share-dialog preview; no public share created",
      "captured_at": "2026-10-29T10:25:42.334Z",
      "status": "partial",
      "body_completeness": "partial",
      "answer_chars": 2417,
      "source_count": 2,
      "notes": "The issue recurred in a new conversation under the same conditions on 2026-10-29. The page indicated that the answer had completed, but the main answer text could not load. The share preview recovered 2417 characters and 2 direct links, ending at Peer-reviewed in the second study. The UI alone cannot establish a server-side generation failure as the cause. The answer did not pass the completeness check and is excluded from the completed sample.",
      "capture_kind": "share_preview",
      "selected": true,
      "file": "data/captures/C-Q2-R1-A3-recovery-01.json",
      "public_evidence_scope": "Public capture summary; original screenshots, full HTML and private conversation information remain local."
    }
  ],
  "capture_provenance": {
    "original_status": "incomplete",
    "selected_capture_id": "C-Q2-R1-A3-recovery-01",
    "method": "DOM-backed share-dialog preview; no public share created",
    "captured_at": "2026-10-29T10:25:42.334Z",
    "same_conversation": true,
    "review_notes": "The issue recurred in a new conversation under the same conditions on 2026-10-29. The page indicated that the answer had completed, but the main answer text could not load. The share preview recovered 2417 characters and 2 direct links, ending at Peer-reviewed in the second study. The UI alone cannot establish a server-side generation failure as the cause. The answer did not pass the completeness check and is excluded from the completed sample.",
    "selected_capture_file": "data/records/C-Q2-R1-A3.json"
  },
  "public_copy": true
}
