As AI eats the web, the internet's collective memory is disappearing

When search invents the facts

Google's AI summaries have begun producing false answers to simple questions. Sunset chasers noticed the feature inventing times: one user in Colorado Springs wrote on Facebook that he had set up a projector outside and was waiting for the sun to set, but "AI informed me the sunset had already happened." The errors are minor, writes Vass Bednar in The Walrus, but they point to a bigger problem: the world's dominant search engine now struggles over basic facts.

Bednar argues the issue is not just quality but memory. Link rot erases pages daily — key sections of the United States Constitution briefly vanished from the Library of Congress website because of a coding error. When an error-prone AI sits between users and an original source, that source can become practically undiscoverable even if it still exists.

Archives are disappearing

Wikipedia faces the opposite pressure: AI systems scrape and ingest its articles, so users no longer click through, and traffic, attention and donations no longer flow back to the encyclopedia. The Internet Archive, whose Wayback Machine has captured hundreds of billions of web snapshots, is straining under storage and engineering demands, cyberattacks and costly litigation. After publishers won a suit over its digital lending programme, some news organizations began blocking the Wayback Machine's crawlers, fearing the archive gives AI companies indirect access to copyrighted material.

Governments look for alternatives

France started treating foreign platforms as a threat to technological sovereignty years ago: its national assembly and armed forces ministry began adopting Qwant, a French search engine hosted in Europe that does not store search data or sell personal information, and the European Parliament, with its 720 members, followed. French civil servants are required to use Tchap, a homegrown messaging app. Across Europe, governments are replacing Microsoft products with open-source tools on hundreds of thousands of workstations, and the European Commission has announced it will join W Social, an independent social site built by a Swedish start-up. "Far too much public digital infrastructure is today tied up with very few foreign suppliers," Denmark's digitalization minister, Caroline Stage Olsen, said.

A court ruling and the case for public infrastructure

In a ruling that could set a precedent, a German court held Google liable for false statements generated by its AI overview feature after the tool wrongly linked two publishing companies to scammy business practices. Because the search engine rewrites information in its own words, the court reasoned, it is doing more than pointing users to the public record — it is authoring a new layer of content, and with that comes editorial responsibility. Google is appealing.

Bednar argues that preserving the record cannot remain a side project. Canada already has pieces of such an architecture — Library and Archives Canada, Internet Archive Canada, the CANARIE research network and the Public Knowledge Project, which has published Open Journal Systems since 2002.