{
  "reviewed": "2026-10-06",
  "method": "Purposive study-comparability audit, not systematic review or meta-analysis",
  "records": [
    {
      "study": "Liu Zhang Liang 2023",
      "source": "https://aclanthology.org/2023.findings-emnlp.467/",
      "publication": "peer-reviewed conference findings",
      "unit": "generated sentence",
      "metric": "fully supported by citations",
      "reported_percent": 51.5,
      "scope": "four historical generative search engines",
      "pool_with_other_rows": false
    },
    {
      "study": "Liu Zhang Liang 2023",
      "source": "https://aclanthology.org/2023.findings-emnlp.467/",
      "unit": "supplied citation",
      "metric": "supports associated sentence",
      "reported_percent": 74.5,
      "scope": "same study, different denominator",
      "pool_with_other_rows": false
    },
    {
      "study": "Tow Center 2025",
      "source": "https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php",
      "publication": "research-centre audit",
      "unit": "news excerpt identification query",
      "metric": "incorrect answer",
      "reported_bound": "greater than 60 percent",
      "queries": 1600,
      "tools": 8,
      "publishers": 20,
      "articles_per_publisher": 10,
      "pool_with_other_rows": false
    },
    {
      "study": "Liu et al 2026",
      "source": "https://arxiv.org/abs/2608.13786",
      "publication": "preprint",
      "unit": "response-level recall of expert-included studies",
      "metric": "mean recall",
      "reported_percent": 39.2,
      "responses": 720,
      "review_questions": 20,
      "models": 3,
      "roles": 3,
      "repetitions": 4,
      "pool_with_other_rows": false
    }
  ],
  "exclusions": "No fabricated 10,000-citation experiment; no overall accuracy ranking; no pooling across units, years or domains"
}