Google AI Overviews hallucinate sunset times as link rot and archive deletions accelerate

AI Overviews now sit between queries and source pages. User reports document fabricated twilight times that shifted local event planning. The Walrus article records these failures alongside upstream content injection on Reddit aimed at shaping model outputs. FiveThirtyEight's full archive vanished after Disney ownership ended in March 2025 with no public index remaining. Link rot studies show 20-30 percent of web pages from 2013 unreachable by 2023. Wikipedia traffic from search engines dropped as AI scrapers consumed its text without returning referral volume. Disney's deletion of a decade of polling data illustrates how commercial asset decisions override archival continuity. Digital erasure differs from physical book bans because no physical copy persists once the host removes the record. Search engines that once indexed stable corpora now train on contaminated, decaying inputs. Canadian content regulations address distribution but leave preservation and retrieval functions unaddressed. Retrieval failure rates will rise as training corpora incorporate more synthetic text. Primary sources become practically inaccessible when AI layers return confident errors instead of links.