保存尝试显示出规模与筛选的限制。2010 年，Library of Congress 认为 Twitter 是当代历史的重要遗产并试图封存每一则推文，但到 2017 年它改为仅保存少量选定贴文。荷兰顾问 Maurice de Kunder 持续追踪被搜寻引擎索引的网页数，发现从 4.7bn 降到 3.98bn。Elon Musk 的 Department of Government Efficiency 据称计划移除美国联邦网站中高达 20% 的内容；随后多家大型公司改写自家网站，删除气候变迁相关文字，这些变更主要也是由第三方才被纪录到。
对数位记忆的依赖仍然零散。虽然实体馆藏仍可保留 1665 年开始的报纸版本，但像威尔士 The National 这类近年新闻站点可在短期内下线或变成 404。Wayback Machine（American non-profit Internet Archive 的方案）已撷取超过 1tn 快照，且 archive.today 也提供补充，但两者都无法保证长久：版权移除、封锁、对 archive.today 背后注册商的 FBI 传票，以及 2025 年下半年快照量下滑都显示其脆弱性。AI 训练可能保留内容痕迹，但难以还原原始形态，因此文中认为有价值的内容仍必须由人为主动即时保存。
The article argues that the internet’s apparent vastness does not mean permanence. A major 2013 web-page survey found that over one third of pages available then were no longer reachable, showing “link rot” as a persistent trend. Viral AI content such as the cartoon-fruit Love Island videos can gain millions of views, yet not everything online deserves preservation; in a web where posts disappear or are quickly replaced, separating what deserves collective cultural memory from what is disposable is becoming harder.
Preservation efforts reveal limits of scale and selection. In 2010, the Library of Congress treated Twitter as an important modern-historical legacy and tried to archive every tweet, but by 2017 it had shifted to saving only a small number of selected posts. Dutch consultant Maurice de Kunder, tracking pages indexed by search engines, found numbers fall from 4.7bn to 3.98bn. Last year, Elon Musk’s Department of Government Efficiency reportedly planned to remove up to 20% of U.S. federal websites; months later, major companies rewrote their sites to remove climate-change references, and those changes were mainly detected by third parties.
Reliance on digital memory remains fragmented. Physical archives still preserve newspapers from as early as 1665, yet modern sites can disappear and return 404s quickly, as with Welsh outlet The National. The Wayback Machine, run by American non-profit Internet Archive, has captured over 1tn snapshots and archive.today provides another route, but neither is guaranteed: copyright removals, blocklisting, an FBI subpoena of archive.today’s registrar, and a 2025 second-half dip in snapshots all indicate fragility. AI training may retain traces of content, but does not reliably preserve original form, so the article concludes that material of value must be intentionally archived by people before it vanishes.