{
 "question": "Where in the path from results page to AI Overview do cited sources come from: how often does an organic page-one result get cited, how steeply does that depend on its position, and where do the other citations come from?",
 "population": "All 4,051 distinct cited URLs in the 481 AI Overviews (with at least one citation) from the 800-keyword US sample, and all 4,016 organic results on those 481 results pages.",
 "inclusion": "URLs in the AI Overview's reference list, per-paragraph references and inline links, de-duplicated after normalization (tracking parameters, fragments and www removed).",
 "exclusion": "Question-keyword supplement (kept for study 4 only).",
 "calculations": [
  "Exact match = normalized URL identical to a page-one organic URL. Looser match = same host (m./amp. removed) and lowercased path without query string, index file, .html or AMP suffix.",
  "Elsewhere on the results page = the URL appears in a non-organic results-page feature (video carousel, images, perspectives, short videos, forums, products, local pack).",
  "Mutually exclusive classes, applied in order: exact organic, looser organic, elsewhere on the page, Google's own pages, other page on a page-one domain, domain not in page-one organic results.",
  "Citation rate by position = organic results at that position that the AI Overview cites / results pages that had an organic result at that position; 95% Wilson intervals.",
  "Citation-level shares: 95% intervals by bootstrap over AI Overviews (2,000 resamples, seed 5), because citations in one AI Overview are not independent.",
  "Position model: logistic regression, organic result cited ~ position + industry + number of citations in the AI Overview, standard errors clustered by AI Overview."
 ],
 "limitations": [
  "Associations, not causes: the study observes what was cited, not how Google retrieved or ranked candidates.",
  "One results page per keyword on one day; run-to-run and week-to-week stability are not measured here.",
  "Page-one organic results only (a median of 9 per page, not 10); a cited page may rank lower.",
  "Citation is measured, not how much of the answer each source supplied.",
  "google.com links are Google's own pages (shopping, local listings) and cannot be organic results; they are a separate class."
 ],
 "update": "monthly",
 "@context": "https://schema.org",
 "@type": "Dataset",
 "name": "Do AI Overviews cite the pages that rank? 4,051 citations",
 "version": "1.1",
 "dateCreated": "2026-09-26",
 "creator": {
  "@type": "Organization",
  "name": "Underneath",
  "url": "https://underneath.agency"
 },
 "url": "https://underneath.agency/research/ai-overview-citations-study",
 "temporalCoverage": "2026-09-26",
 "isAccessibleForFree": true,
 "license": "https://creativecommons.org/licenses/by/4.0/",
 "code": "cite/pipeline/s5_v11.py, s5_v11_package.py (version 1.1); analyze_serp.py (version 1.0)",
 "variableMeasured": [
  "cited_url",
  "cited_domain",
  "source_type",
  "match_class",
  "organic_position",
  "organic_results_on_page",
  "industry"
 ],
 "dateModified": "2026-09-28"
}