{
 "question": "Which sources do AI engines cite in time, and why: do fresher pages win within the same question or search (selection preference), do the questions asked call for recent pages (query-driven recency), or do engines cite kinds of pages that are recent anyway (source composition)? And is the recency in newly published pages or in recent modified dates?",
 "population": "Version 1.0 sample (up to 400 cited pages per engine, seed 20260926, platform pages and PDFs excluded) linked back to the queries that cited them; Google top 10 for 485 searches (study 10); new in 1.1: Google top 10 for the assistants' own 80 questions (study 23, 26 September 2026), 547 non-platform pages fetched by plain HTTP on 28 September 2026.",
 "inclusion": "Pages fetched by plain HTTP with a machine-readable date.",
 "exclusion": "Pages that could not be fetched or carried no date; platform pages (YouTube, Reddit, social networks, Google, Amazon, Yelp) and PDFs.",
 "calculations": [
  "Age = days from the newest machine-readable date to 26 September 2026 (the version 1.0 rule). Publish age = days from the earliest declared publish date (JSON-LD datePublished or uploadDate, article:published_time). New page = publish date within 90 days.",
  "Date anatomy of each dated page: new page; older page with a recent (90-day) modified date; recent date with no publish date; newest date within one day of the fetch (likely stamped automatically); not updated in 90 days.",
  "Same-question comparison: cited pages against Google top-10 pages for the same assistant question, question fixed effects; outcomes log(1 + age) (reported as a ratio of geometric means), dated within 90 days, new page, log(1 + publish age).",
  "Selection within the same results: linear probability models of being cited with query fixed effects and dummies for organic positions 2 to 10; assistants: Google top-10 pages for the question, stacked over the four engines; AI Overviews: the study 10 top-10 pages. Split by DataForSEO intent and by query temporal class.",
  "Query temporal class (rules on the query text): time-sensitive (a year, latest, new, current, now, price, cost, rates, cheap, fees, deals and similar), recommendation (best, top, which, reviews, compare, reliable and similar), other.",
  "Composition: page kind = article markup (yes or no) by source group (reference, review site or directory, retailer, other website); each engine's dated pages reweighted to the Google top-10 kind mix (direct standardization).",
  "Intervals: 2,000 bootstrap resamples of whole queries (1,000 for models and new-page shares), seed 20260928; percentile 95% intervals."
 ],
 "limitations": [
  "Observational: nothing was changed on any page, so none of these differences is a causal effect of freshness.",
  "Retrieval is not observed: we see which pages were cited, and Google's top 10 as a stand-in for what could have been found.",
  "Declared dates are what a page says, not what changed on it; no content was compared across versions.",
  "One collection date; run-to-run and month-to-month stability are not measured.",
  "71 of the 80 assistant questions are recommendation questions (best, which) and only 9 are time-sensitive, too few to test query temporality for the assistants.",
  "Same-question Google pages were fetched two days after the cited pages; any page updated in between would look slightly fresher, which narrows the measured gap.",
  "The baseline pages of study 10 were also fetched with Firecrawl when blocked; the cited pages and the same-question pages were not."
 ],
 "update": "quarterly",
 "@context": "https://schema.org",
 "@type": "Dataset",
 "name": "How fresh are the pages AI engines cite?",
 "version": "1.1",
 "dateCreated": "2026-09-26",
 "creator": {
  "@type": "Organization",
  "name": "Underneath",
  "url": "https://underneath.agency"
 },
 "url": "https://underneath.agency/research/ai-source-freshness-study",
 "temporalCoverage": "2026-09-26",
 "isAccessibleForFree": true,
 "license": "https://creativecommons.org/licenses/by/4.0/",
 "code": "cite/pipeline/s22_v11.py (version 1.1: collect, analyze, package); s22_fresh.py (version 1.0)",
 "variableMeasured": [
  "engine",
  "query",
  "age_days",
  "publish_age_days",
  "new_page",
  "date_anatomy",
  "article_markup",
  "source_group",
  "query_temporal_class",
  "intent",
  "organic_position",
  "cited"
 ],
 "dateModified": "2026-09-28"
}